File size: 138,883 Bytes
496ab4f
e5187e3
37daf11
 
 
e5187e3
b6d3026
e5187e3
 
 
37daf11
496ab4f
e5187e3
 
496ab4f
 
e5187e3
 
 
 
 
37daf11
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d3026
 
 
e5187e3
496ab4f
e5187e3
496ab4f
e5187e3
496ab4f
e5187e3
496ab4f
e5187e3
 
 
 
 
37daf11
 
 
b6d3026
37daf11
b6d3026
 
37daf11
496ab4f
e5187e3
 
 
37daf11
 
 
 
 
 
 
 
 
 
 
 
 
 
496ab4f
 
 
e5187e3
496ab4f
37daf11
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
496ab4f
e5187e3
 
 
 
37daf11
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
496ab4f
e5187e3
 
 
 
 
 
 
 
 
37daf11
 
496ab4f
e5187e3
 
496ab4f
 
 
 
37daf11
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
{"slug":"tml-interaction-small","name":"TML-Interaction-Small","vendor":"Thinking Machines Lab","tagline":"276B-A12B mixture-of-experts interaction model with VAD-free, time-aligned 200 ms micro-turns across audio, video, and text — Thinking Machines Lab's first public voice release.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-05","first_response_ms":400,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"preview","benchmarks":["fdb-v15","audio-multichallenge"],"url":"https://thinkingmachines.ai/blog/interaction-models/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/tml-interaction-small"}
{"slug":"moshi","name":"Moshi","vendor":"Kyutai","tagline":"A 7B speech-text foundation model that generates user and assistant audio streams in parallel.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-09","first_response_ms":160,"open_weights":true,"license":"Apache-2.0 (code) / CC-BY-4.0 (weights)","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":["fdb-v1","big-bench-audio","talking-turns","duplexact","turnbench"],"url":"https://kyutai.org/blog/2024-09-18-moshi-release/","paper_url":"https://arxiv.org/abs/2410.00037","repo_url":"https://github.com/kyutai-labs/moshi","page_url":"https://www.fullduplex.ai/models/moshi"}
{"slug":"j-moshi","name":"J-Moshi / J-Moshi-ext","vendor":"Nagoya University (Higashinaka Lab)","tagline":"Kyutai's 7B Moshi adapted to Japanese, with overlaps and backchannels in real time; the authors present it as the first publicly available Japanese full-duplex dialogue model.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-01","first_response_ms":null,"open_weights":true,"license":"CC-BY-NC-4.0","commercial_use":"non-commercial","languages":["ja"],"status":"current","benchmarks":["j-moshi","m3-duplexbench"],"url":"https://nu-dialogue.github.io/j-moshi","paper_url":"https://arxiv.org/abs/2506.02979","repo_url":"https://github.com/nu-dialogue/j-moshi","page_url":"https://www.fullduplex.ai/models/j-moshi"}
{"slug":"llm-jp-moshi-v1","name":"LLM-jp-Moshi-v1","vendor":"NII (LLM-jp)","tagline":"Japanese full-duplex spoken dialogue model on the 7B Moshi architecture, released by NII under Apache-2.0 so it can be used commercially.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-02","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["ja"],"status":"current","benchmarks":["m3-duplexbench"],"url":"https://llm-jp.github.io/llm-jp-moshi/","paper_url":null,"repo_url":"https://github.com/llm-jp/llm-jp-moshi","page_url":"https://www.fullduplex.ai/models/llm-jp-moshi-v1"}
{"slug":"shizuku-duplex","name":"Shizuku-Duplex","vendor":"Shizuku AI","tagline":"Shizuku AI's Japanese full-duplex model for its AI character: it listens and speaks on an 80 ms tick and hands web search and tool calls to a background model. No weights or API have been released.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["ja"],"status":"preview","benchmarks":[],"url":"https://shizuku.ai/en/blog/1/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/shizuku-duplex"}
{"slug":"glm-4-voice","name":"GLM-4-Voice","vendor":"Zhipu AI (Z.ai)","tagline":"An end-to-end Chinese/English speech LM that exposes tone, dialect, and emotion as instructions.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-10","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code) / GLM-4 licence (weights; commercial use after registration)","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/zai-org/GLM-4-Voice","paper_url":"https://arxiv.org/abs/2412.02612","repo_url":"https://github.com/zai-org/GLM-4-Voice","page_url":"https://www.fullduplex.ai/models/glm-4-voice"}
{"slug":"sesame-csm","name":"Sesame CSM","vendor":"Sesame AI","tagline":"The conversational speech model everyone called \"uncanny-valley-free.\"","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-03","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice","paper_url":null,"repo_url":"https://github.com/SesameAILabs/csm","page_url":"https://www.fullduplex.ai/models/sesame-csm"}
{"slug":"kimi-audio","name":"Kimi-Audio","vendor":"Moonshot AI","tagline":"A 7B audio foundation model covering ASR, AQA, audio QA, and end-to-end voice chat in one net.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-04","first_response_ms":null,"open_weights":true,"license":"MIT (Apache-2.0 for code derived from Qwen2.5-7B)","commercial_use":"yes","languages":["zh","en"],"status":"current","benchmarks":["voicebench","mmar","air-bench"],"url":"https://github.com/MoonshotAI/Kimi-Audio","paper_url":"https://arxiv.org/abs/2504.18425","repo_url":"https://github.com/MoonshotAI/Kimi-Audio","page_url":"https://www.fullduplex.ai/models/kimi-audio"}
{"slug":"personaplex","name":"PersonaPlex","vendor":"NVIDIA (ADLR)","tagline":"7B full-duplex speech LM with hybrid voice + text prompts — any role, any voice, on top of Moshi.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-01","first_response_ms":170,"open_weights":true,"license":"MIT (code) / NVIDIA Open Model License (weights)","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":["fdb-v1"],"url":"https://research.nvidia.com/labs/adlr/personaplex/","paper_url":"https://arxiv.org/abs/2602.06053","repo_url":"https://github.com/NVIDIA/PersonaPlex","page_url":"https://www.fullduplex.ai/models/personaplex"}
{"slug":"minicpm-o-4-5","name":"MiniCPM-o 4.5","vendor":"OpenBMB / ModelBest","tagline":"9B on-device omni-modal LLM with full-duplex speech + vision streaming, targeting Gemini 2.5 Flash quality on a laptop.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-02","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh","+30"],"status":"current","benchmarks":["videofdb","duplexact","multipartyturnbench"],"url":"https://github.com/OpenBMB/MiniCPM-o","paper_url":"https://arxiv.org/abs/2604.27393","repo_url":"https://github.com/OpenBMB/MiniCPM-o","page_url":"https://www.fullduplex.ai/models/minicpm-o-4-5"}
{"slug":"salmonn-omni","name":"SALMONN-omni","vendor":"ByteDance · Tsinghua · Cambridge","tagline":"Standalone codec-free full-duplex speech LLM with no audio tokens in its vocabulary, trained with reinforcement learning for barge-in.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2505.17060","paper_url":"https://arxiv.org/abs/2505.17060","repo_url":null,"page_url":"https://www.fullduplex.ai/models/salmonn-omni"}
{"slug":"step-audio-2-mini","name":"Step-Audio 2 mini","vendor":"StepFun","tagline":"End-to-end multimodal audio LLM for speech conversation with expressive TTS and tool / search integration.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-08","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["zh","en"],"status":"current","benchmarks":["big-bench-audio","voicebench","uro-bench"],"url":"https://github.com/stepfun-ai/Step-Audio2","paper_url":"https://arxiv.org/abs/2507.16632","repo_url":"https://github.com/stepfun-ai/Step-Audio2","page_url":"https://www.fullduplex.ai/models/step-audio-2-mini"}
{"slug":"covo-audio-chat","name":"Covo-Audio-Chat / Covo-Audio-Chat-FD","vendor":"Tencent","tagline":"7B end-to-end audio LLM with a native full-duplex variant — tri-modal speech-text interleaving and dual-stream listen-and-speak.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-03","first_response_ms":null,"open_weights":true,"license":"Covo-Audio licence (academic use only)","commercial_use":"research-only","languages":["zh","en"],"status":"preview","benchmarks":["uro-bench","voicebench"],"url":"https://github.com/Tencent/Covo-Audio","paper_url":"https://arxiv.org/abs/2602.09823","repo_url":"https://github.com/Tencent/Covo-Audio","page_url":"https://www.fullduplex.ai/models/covo-audio-chat"}
{"slug":"llama-omni2","name":"LLaMA-Omni 2","vendor":"ICT / CAS (ictnlp)","tagline":"Qwen2.5-based real-time spoken chatbot series (0.5B–14B) with autoregressive streaming speech synthesis.","category":"speech-lm-fd","duplex_type":"half-duplex","full_duplex":false,"released":"2025-04","first_response_ms":583,"open_weights":true,"license":"Apache-2.0 (code) / academic research only (weights)","commercial_use":"research-only","languages":["en"],"status":"current","benchmarks":["voicebench"],"url":"https://github.com/ictnlp/LLaMA-Omni2","paper_url":"https://arxiv.org/abs/2505.02625","repo_url":"https://github.com/ictnlp/LLaMA-Omni2","page_url":"https://www.fullduplex.ai/models/llama-omni2"}
{"slug":"realtime-venus","name":"Realtime-Venus","vendor":"Ant Group (inclusionAI)","tagline":"Two open 9B full-duplex models, one audio-visual and one speech-only, that keep talking while a background harness runs reasoning and tools.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh"],"status":"current","benchmarks":["fdb-v15"],"url":"https://realtime-venus.github.io/","paper_url":"https://arxiv.org/abs/2609.13814","repo_url":"https://github.com/inclusionAI/Realtime-Venus","page_url":"https://www.fullduplex.ai/models/realtime-venus"}
{"slug":"duplexomni","name":"DuplexOmni","vendor":"Xi'an Jiaotong University / Peking University / Meituan","tagline":"Open full-duplex omni model that listens and watches in 480 ms slices, answers in speech in real time, and hands hard questions to a pluggable thinking layer.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-07","first_response_ms":506,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["zh","en"],"status":"preview","benchmarks":["big-bench-audio"],"url":"https://huggingface.co/MuyeHuang/DuplexOmni","paper_url":"https://arxiv.org/abs/2606.09186","repo_url":"https://github.com/MuyeHuang/DuplexOmni","page_url":"https://www.fullduplex.ai/models/duplexomni"}
{"slug":"nemotronlabs-voicechat","name":"NVIDIA NemotronLabs VoiceChat 11B","vendor":"NVIDIA","tagline":"Open 11B full-duplex speech model on a Nemotron Nano v2 backbone, with a separate output channel for tool calls so it keeps talking while tools run.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-08","first_response_ms":448,"open_weights":true,"license":"OpenMDW-1.1","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":["fdb-v1","voicebench","fdb-v3"],"url":"https://huggingface.co/nvidia/NVIDIA-NemotronLabs-VoiceChat-11B","paper_url":"https://arxiv.org/abs/2609.21967","repo_url":"https://github.com/NVIDIA-NeMo/Speech/tree/nemotron-labs-voicechat","page_url":"https://www.fullduplex.ai/models/nemotronlabs-voicechat"}
{"slug":"raon-speechchat","name":"Raon-SpeechChat-9B","vendor":"KRAFTON","tagline":"Open 9B English full-duplex speech model with a dedicated backchannel token, speak-first or listen-first modes, and voice conditioning.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"CC-BY-NC-4.0","commercial_use":"non-commercial","languages":["en"],"status":"current","benchmarks":["duplexact"],"url":"https://huggingface.co/KRAFTON/Raon-SpeechChat-9B","paper_url":"https://arxiv.org/abs/2605.23912","repo_url":"https://github.com/krafton-ai/Raon-Speech","page_url":"https://www.fullduplex.ai/models/raon-speechchat"}
{"slug":"lychee-fd","name":"Lychee-FD","vendor":"HITsz-TMG","tagline":"Open full-duplex speech model that keeps listening, understanding, speaking and interaction control in one model, with acoustic and semantic information modelled in separate layers.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-07","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["zh","en"],"status":"current","benchmarks":["duplexact"],"url":"https://github.com/HITsz-TMG/Lychee-FD","paper_url":"https://arxiv.org/abs/2607.06540","repo_url":"https://github.com/HITsz-TMG/Lychee-FD","page_url":"https://www.fullduplex.ai/models/lychee-fd"}
{"slug":"gander","name":"Gander","vendor":"Omni-Interaction-Gander","tagline":"Open multimodal duplex interaction model built on MiniCPM-o 4.5 that listens and speaks while also taking in images and video.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/Omni-Interaction-Gander/Omni-Interaction-Agent","paper_url":"https://arxiv.org/abs/2609.08977","repo_url":"https://github.com/Omni-Interaction-Gander/Omni-Interaction-Agent","page_url":"https://www.fullduplex.ai/models/gander"}
{"slug":"bayling-duplex","name":"BayLing-Duplex","vendor":"ICT, Chinese Academy of Sciences","tagline":"Native full-duplex speech dialogue in a single autoregressive LLM, with turn-taking handled by dialogue-state tokens instead of a VAD.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-06","first_response_ms":null,"open_weights":true,"license":"glm-4-voice License (commercial use after registration)","commercial_use":"yes","languages":[],"status":"current","benchmarks":["duplexact"],"url":"https://github.com/BayLing-Models/BayLing-Duplex","paper_url":"https://arxiv.org/abs/2606.14528","repo_url":"https://github.com/BayLing-Models/BayLing-Duplex","page_url":"https://www.fullduplex.ai/models/bayling-duplex"}
{"slug":"ellsa","name":"ELLSA","vendor":"Tsinghua University / ByteDance","tagline":"Open full-duplex model that takes speech and camera frames and produces speech and robot actions in one streaming loop, tested in the LIBERO simulator.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code and weights) on a Llama-3.1-8B-Instruct base (llama3.1 licence)","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/bytedance/SALMONN/tree/ELLSA","paper_url":"https://arxiv.org/abs/2510.16756","repo_url":"https://github.com/bytedance/SALMONN/tree/ELLSA","page_url":"https://www.fullduplex.ai/models/ellsa"}
{"slug":"flm-audio","name":"FLM-Audio","vendor":"CofeAI (BAAI)","tagline":"A Chinese and English full-duplex model on a 7B backbone that listens, speaks and writes a sentence-level inner monologue in the same time steps.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-07","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (repository) / research use only (README and Hub notice)","commercial_use":"research-only","languages":["zh","en"],"status":"current","benchmarks":["librispeech","seed-tts-eval"],"url":"https://github.com/cofe-ai/flm-audio","paper_url":"https://arxiv.org/abs/2509.02521","repo_url":"https://github.com/cofe-ai/flm-audio","page_url":"https://www.fullduplex.ai/models/flm-audio"}
{"slug":"haru-4b","name":"Haru-4B","vendor":"higurashi","tagline":"Open full-duplex base model built on Qwen3.5-4B, in nine languages, that takes in and produces one 133 ms frame of continuous audio per step.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-11","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","es","pt","de","fr","zh","ja","ru","ko"],"status":"current","benchmarks":[],"url":"https://blog.higurashi.ai/melte/haru4b/","paper_url":null,"repo_url":"https://github.com/higurashi-ai/Haru","page_url":"https://www.fullduplex.ai/models/haru-4b"}
{"slug":"hertz-dev","name":"Hertz-dev","vendor":"Standard Intelligence","tagline":"An 8.5B audio-only base model for full-duplex conversation that predicts audio latents for two speaker channels at 8 Hz.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-11","first_response_ms":120,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://si.inc/posts/hertz-dev","paper_url":null,"repo_url":"https://github.com/Standard-Intelligence/hertz-dev","page_url":"https://www.fullduplex.ai/models/hertz-dev"}
{"slug":"human-1","name":"Human-1","vendor":"Josh Talks","tagline":"Hindi full-duplex dialogue model adapted from Moshi with a Hindi tokeniser and trained on 26,000 hours of two-channel conversation.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-03","first_response_ms":2021,"open_weights":true,"license":"CC-BY-4.0","commercial_use":"yes","languages":["hi"],"status":"current","benchmarks":[],"url":"https://huggingface.co/VoiceArena/Human-1","paper_url":"https://arxiv.org/abs/2604.23295","repo_url":null,"page_url":"https://www.fullduplex.ai/models/human-1"}
{"slug":"kame","name":"KAME","vendor":"Sakana AI","tagline":"Moshi-based speech-to-speech front end that keeps talking while a back-end text LLM sends it candidate replies through an added oracle stream.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"MIT (code) / not stated (weights)","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://pub.sakana.ai/kame/","paper_url":"https://arxiv.org/abs/2510.02327","repo_url":"https://github.com/SakanaAI/kame","page_url":"https://www.fullduplex.ai/models/kame"}
{"slug":"voila","name":"Voila (Voila-chat / Voila-autonomous preview)","vendor":"Maitrix.org","tagline":"Open voice-language model family with an end-to-end chat model and a preview full-duplex model that processes the user's audio stream and its own at once.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-04","first_response_ms":null,"open_weights":true,"license":"MIT (code and Hub tag); backbone LLM not named","commercial_use":"unclear","languages":["en","zh","fr","de","ja","ko"],"status":"preview","benchmarks":["librispeech"],"url":"https://voila.maitrix.org","paper_url":"https://arxiv.org/abs/2505.02707","repo_url":"https://github.com/maitrix-org/Voila","page_url":"https://www.fullduplex.ai/models/voila"}
{"slug":"egovoice","name":"EgoVoice","vendor":"University of Seoul","tagline":"Proactive spoken assistant that watches first-person video and audio and decides each second whether to stay silent or speak guidance.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://egocentricvoice.github.io/","paper_url":"https://arxiv.org/abs/2610.12248","repo_url":null,"page_url":"https://www.fullduplex.ai/models/egovoice"}
{"slug":"adaptduplex","name":"AdaptDuplex","vendor":"Zuoyebang","tagline":"Full-duplex model built on Qwen3-Omni that picks the length of its next listening window and exposes each behaviour decision as a token.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["fdb-v1","fdb-v15","humdial"],"url":"https://adaptduplex.github.io/","paper_url":"https://arxiv.org/abs/2609.29217","repo_url":null,"page_url":"https://www.fullduplex.ai/models/adaptduplex"}
{"slug":"charduplex","name":"CharDuplex","vendor":"Nanyang Technological University / AI Singapore / UC Berkeley","tagline":"Full-duplex speech model adapted from GLM-4-Voice that stays in a prescribed character, trained on character dialogues and then by reinforcement learning against a simulated user.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["fdb-v1","voicebench"],"url":"https://arxiv.org/abs/2609.34461","paper_url":"https://arxiv.org/abs/2609.34461","repo_url":null,"page_url":"https://www.fullduplex.ai/models/charduplex"}
{"slug":"counselplex","name":"CounselPlex","vendor":"KAIST (Integrated Vision and Language Lab)","tagline":"PersonaPlex fine-tuned for emotional-support conversation, which writes a short note on the seeker's stressor during silences and re-injects it while it listens.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-05","first_response_ms":null,"open_weights":true,"license":"MIT (code) / no licence stated for the weights","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["fdb-v1"],"url":"https://github.com/yeonju7kim/CounselPlex","paper_url":"https://arxiv.org/abs/2610.32702","repo_url":"https://github.com/yeonju7kim/CounselPlex","page_url":"https://www.fullduplex.ai/models/counselplex"}
{"slug":"dgslm","name":"dGSLM","vendor":"Meta AI","tagline":"Textless model that generates both channels of a spoken dialogue, with overlaps and laughter, from discrete units learned on Fisher telephone calls.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2023-01","first_response_ms":null,"open_weights":true,"license":"MIT (fairseq code) / no licence stated for the checkpoints","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/facebookresearch/fairseq/tree/main/examples/textless_nlp/dgslm","paper_url":"https://arxiv.org/abs/2203.16502","repo_url":"https://github.com/facebookresearch/fairseq/tree/main/examples/textless_nlp/dgslm","page_url":"https://www.fullduplex.ai/models/dgslm"}
{"slug":"diffuplex","name":"Diffuplex","vendor":"NVIDIA","tagline":"Full-duplex spoken language model in which an autoregressive policy decides when to speak and a discrete diffusion model writes the reply in parallel.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":396,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["voicebench","fdb-v1","fdb-v3","big-bench-audio"],"url":"https://cchen1436.github.io/diffuplex-website/","paper_url":"https://arxiv.org/abs/2610.17780","repo_url":null,"page_url":"https://www.fullduplex.ai/models/diffuplex"}
{"slug":"duplexsla","name":"DuplexSLA","vendor":"StepFun / Peking University / Nanyang Technological University","tagline":"7B full-duplex model built on Step-Audio 2 mini that decodes assistant speech and a text action channel for planning and tool calls on one 160 ms clock.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-05","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://hyzhang24.github.io/DuplexSLA/","paper_url":"https://arxiv.org/abs/2605.20755","repo_url":"https://github.com/hyzhang24/DuplexSLA","page_url":"https://www.fullduplex.ai/models/duplexsla"}
{"slug":"flair","name":"FLAIR","vendor":"Nanyang Technological University / Mila","tagline":"Full-duplex spoken dialogue model that reasons in latent embeddings while the user is still speaking, with no separate thinking phase before it answers.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["voicebench","fdb-v1"],"url":"https://arxiv.org/abs/2603.17837","paper_url":"https://arxiv.org/abs/2603.17837","repo_url":null,"page_url":"https://www.fullduplex.ai/models/flair"}
{"slug":"diffuplex-listen-speak","name":"DiffuPlex-LISTEN / DiffuPlex-SPEAK","vendor":"University of Seoul","tagline":"Rolling masked diffusion add-on for Kyutai's personaplex-rl-seamless that predicts several future user and assistant frames per backbone call to cut sequential computation.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["fdb-v1","fdb-v15","voicebench"],"url":"https://diffuplex.github.io/","paper_url":"https://arxiv.org/abs/2610.12214","repo_url":null,"page_url":"https://www.fullduplex.ai/models/diffuplex-listen-speak"}
{"slug":"lslm","name":"LSLM (Listening-while-Speaking Language Model)","vendor":"Shanghai Jiao Tong University (X-LANCE) / ByteDance","tagline":"Full-duplex research model that adds a streaming listening channel to a decoder-only TTS model so it can stop speaking when a user interrupts.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://ddlbojack.github.io/LSLM","paper_url":"https://arxiv.org/abs/2408.02622","repo_url":null,"page_url":"https://www.fullduplex.ai/models/lslm"}
{"slug":"omniflatten","name":"OmniFlatten","vendor":"Alibaba (Tongyi Lab)","tagline":"End-to-end full-duplex model that post-trains Qwen2-0.5B on user speech, assistant text and assistant speech chunks flattened into a single token sequence.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["librispeech"],"url":"https://omniflatten.github.io/","paper_url":"https://arxiv.org/abs/2410.17799","repo_url":null,"page_url":"https://www.fullduplex.ai/models/omniflatten"}
{"slug":"replay","name":"RePlay","vendor":"Disney Research Imagineering / BUT Speech@FIT","tagline":"Spoken dialogue system cut down from PersonaPlex that chooses and plays pre-recorded lines instead of generating speech, with turn-taking and retrieval heads on layer 15.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2609.31588","paper_url":"https://arxiv.org/abs/2609.31588","repo_url":null,"page_url":"https://www.fullduplex.ai/models/replay"}
{"slug":"salmonn-duo","name":"SALMONN-duo","vendor":"Tsinghua University / ByteDance","tagline":"Dual-system voice agent in which an always-on full-duplex speech LLM built on SALMONN-omni decides when to hand a request to an asynchronous backend LLM agent.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2609.34247","paper_url":"https://arxiv.org/abs/2609.34247","repo_url":null,"page_url":"https://www.fullduplex.ai/models/salmonn-duo"}
{"slug":"scot","name":"SCoT (Streaming Chain-of-Thought)","vendor":"Carnegie Mellon University / Sony Group","tagline":"Streaming chain-of-thought training for a single-stream duplex spoken dialogue model that predicts a transcript, a text reply and speech for each two-second block.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2510.02066","paper_url":"https://arxiv.org/abs/2510.02066","repo_url":null,"page_url":"https://www.fullduplex.ai/models/scot"}
{"slug":"steerableplex","name":"SteerablePlex","vendor":"University of Illinois Urbana-Champaign et al.","tagline":"PersonaPlex fine-tuned with reinforcement learning to follow text instructions injected mid-conversation, used with an LLM director as a controllable full-duplex user simulator.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["fdb-v1"],"url":"https://arxiv.org/abs/2610.12201","paper_url":"https://arxiv.org/abs/2610.12201","repo_url":null,"page_url":"https://www.fullduplex.ai/models/steerableplex"}
{"slug":"steerduplex","name":"SteerDuplex","vendor":"Scale AI / University of Maryland","tagline":"Moshi-based full-duplex model fine-tuned to follow spoken requests about tone, persona, style and speed, then tuned with two-stage reinforcement learning for interruption timing.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["audio-multichallenge","voicebench","fdb-v2"],"url":"https://arxiv.org/abs/2609.12623","paper_url":"https://arxiv.org/abs/2609.12623","repo_url":"https://github.com/Utkarsh4430/SteerDuplex","page_url":"https://www.fullduplex.ai/models/steerduplex"}
{"slug":"syncllm","name":"SyncLLM (Synchronous LLMs)","vendor":"Meta AI / University of Washington","tagline":"Llama3-8b post-trained to generate both speakers' speech units in interleaved, clock-synchronised chunks, presented at EMNLP 2024 as a full-duplex dialogue agent.","category":"speech-lm-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2024-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://syncllm.cs.washington.edu/","paper_url":"https://arxiv.org/abs/2409.15594","repo_url":null,"page_url":"https://www.fullduplex.ai/models/syncllm"}
{"slug":"openai-realtime","name":"OpenAI Realtime (gpt-realtime-2.1)","vendor":"OpenAI","tagline":"Realtime API reasoning voice model: gpt-realtime-2.1 updates gpt-realtime-2 with better alphanumeric recognition and steadier handling of silence, noise, and interruptions, plus a cheaper mini sibling.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","es","fr","+50"],"status":"current","benchmarks":["big-bench-audio","fdb-v1","audio-multichallenge","scale-voice-showdown","duplexact"],"url":"https://developers.openai.com/api/docs/models/gpt-realtime-2.1","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/openai-realtime"}
{"slug":"gemini-live","name":"Gemini 2.5 Live","vendor":"Google DeepMind","tagline":"Legacy native-audio Live API model; Google shuts it down no earlier than 2026-11-17 and names Gemini 3.8 Live as the replacement.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","+21"],"status":"legacy","benchmarks":["big-bench-audio","talking-turns","videofdb"],"url":"https://ai.google.dev/gemini-api/docs/live","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gemini-live"}
{"slug":"gemini-3-live","name":"Gemini 3.1 Flash Live Preview","vendor":"Google DeepMind","tagline":"Preview Live API snapshot on 3.1 Flash with proactive audio, native thinking, and tighter tool-call latency; Google shuts it down no earlier than 2026-11-17 and names Gemini 3.8 Live as the replacement.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","+21"],"status":"legacy","benchmarks":["big-bench-audio","talking-turns","videofdb","duplexact","turnbench"],"url":"https://ai.google.dev/gemini-api/docs/live","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gemini-3-live"}
{"slug":"nova-2-sonic","name":"Amazon Nova 2 Sonic","vendor":"Amazon / AWS","tagline":"Bedrock-hosted GA speech-to-speech model — polyglot voices, 1M-token context, async tool calls, telephony-ready.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2025-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es","fr","de","it","pt","hi"],"status":"current","benchmarks":[],"url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-sonic.html","paper_url":"https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf","repo_url":null,"page_url":"https://www.fullduplex.ai/models/nova-2-sonic"}
{"slug":"nova-2-5-sonic","name":"Amazon Nova 2.5 Sonic","vendor":"Amazon / AWS","tagline":"Bedrock speech-to-speech model that succeeds Nova 2 Sonic at the same price, with claimed gains in reasoning, instruction following, and tool-calling accuracy.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-sonic/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/nova-2-5-sonic"}
{"slug":"grok-voice-agent","name":"Grok Voice 3.0 (xAI Speech to Speech API)","vendor":"xAI","tagline":"xAI's speech-to-speech API, now on Grok Voice 3.0: 20+ languages, $0.08 per minute, wire-compatible with the OpenAI Realtime API.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es","fr","de","zh","ja","pt","+13"],"status":"current","benchmarks":["duplexact"],"url":"https://docs.x.ai/developers/model-capabilities/audio/voice-agent","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/grok-voice-agent"}
{"slug":"gpt-live-1","name":"GPT-Live 1","vendor":"OpenAI","tagline":"OpenAI's full-duplex voice model: it listens and talks at the same time and hands reasoning and tool calls to a backend model or agent.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-07","first_response_ms":798,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":["fdb-v1","fdb-v15","fdb-v3","tau3-bench"],"url":"https://developers.openai.com/api/docs/models/gpt-live-1","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gpt-live-1"}
{"slug":"gemini-3-8-live","name":"Gemini 3.8 Live","vendor":"Google DeepMind","tagline":"Google's GA Live API voice model: native audio in and out, camera and screen frames as input, and tool calls that run in the background while it keeps talking.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","+94"],"status":"current","benchmarks":["multipartyturnbench"],"url":"https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gemini-3-8-live"}
{"slug":"seedrealtime","name":"SeedRealtime","vendor":"ByteDance Seed","tagline":"ByteDance Seed's audio-visual full-duplex model: it watches a live camera feed, decides when to speak without an external VAD, and runs video calls in the Doubao app.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://seed.bytedance.com/en/blog/seedrealtime-audio-visual-full-duplex-llm-released-toward-omni-modal-natural-interaction","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/seedrealtime"}
{"slug":"eleven-realtime","name":"Eleven Realtime (v1 Mini / v1 Max)","vendor":"ElevenLabs","tagline":"ElevenLabs' speech-to-speech models, each a complete voice agent: Mini for low-cost conversation, Max for reasoning and multi-step tool use, behind an OpenAI-Realtime-compatible API.","category":"assistant-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["90+"],"status":"current","benchmarks":[],"url":"https://elevenlabs.io/blog/realtime","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/eleven-realtime"}
{"slug":"stepaudio-3-realtime","name":"StepAudio 3 Realtime","vendor":"StepFun","tagline":"StepFun's full-duplex realtime voice model on its open platform: the user and the model can speak at once, and it thinks while it speaks.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"preview","benchmarks":[],"url":"https://platform.stepfun.ai/docs/en/guides/models/stepaudio-3-realtime","paper_url":"https://arxiv.org/abs/2609.14005","repo_url":null,"page_url":"https://www.fullduplex.ai/models/stepaudio-3-realtime"}
{"slug":"qwen-audio-realtime","name":"Qwen-Audio Realtime (3.0 / 3.1)","vendor":"Alibaba Cloud","tagline":"Alibaba Cloud Model Studio's duplex speech models for voice conversation, served over WebSocket, AOQ, or WebRTC with system voices and voice cloning.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["zh","en","ja","ko","de","es","fr","id","it","pt","ru","th","vi","ar"],"status":"current","benchmarks":[],"url":"https://www.alibabacloud.com/help/en/model-studio/qwen-audio-realtime-user-guides","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/qwen-audio-realtime"}
{"slug":"qwen-omni-realtime","name":"Qwen-Omni Realtime (Qwen3.8-Omni-Flash-Realtime)","vendor":"Alibaba Cloud","tagline":"Alibaba Cloud Model Studio's realtime omni API: live audio and video in, speech and text out, with function calling and web search.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["113"],"status":"current","benchmarks":[],"url":"https://www.alibabacloud.com/help/en/model-studio/realtime","paper_url":"https://arxiv.org/abs/2609.25611","repo_url":null,"page_url":"https://www.fullduplex.ai/models/qwen-omni-realtime"}
{"slug":"phonic-1","name":"Phonic-1","vendor":"Phonic","tagline":"Phonic's speech-to-speech model for agentic voice tasks, served through its API and agent builder in 51 languages.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["51"],"status":"current","benchmarks":["tau3-bench"],"url":"https://phonic.ai/blog/phonic-1","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/phonic-1"}
{"slug":"higgs-realtime","name":"Higgs Realtime","vendor":"Boson AI","tagline":"Boson AI's audio-native realtime model and API, compatible with the OpenAI Realtime API and speaking 100+ languages.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-08","first_response_ms":700,"open_weights":false,"license":null,"commercial_use":"yes","languages":["100+"],"status":"current","benchmarks":[],"url":"https://www.boson.ai/blog/higgs-realtime","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/higgs-realtime"}
{"slug":"rime-prism","name":"Rime Prism","vendor":"Rime","tagline":"Rime's speech-to-speech model for voice agents: it reasons over the caller's audio and produces text you can check before it is spoken.","category":"assistant-fd","duplex_type":"half-duplex","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://www.rime.ai/prism","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/rime-prism"}
{"slug":"glm-realtime","name":"GLM-Realtime","vendor":"Zhipu AI (Z.ai)","tagline":"Zhipu's hosted audio and video call model: it takes speech, camera or screen frames and text over a WebSocket and answers in speech.","category":"assistant-fd","duplex_type":"half-duplex","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://docs.bigmodel.cn/cn/guide/models/sound-and-video/glm-realtime","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/glm-realtime"}
{"slug":"smallest-hydra","name":"Hydra","vendor":"Smallest AI","tagline":"Smallest AI's full-duplex speech-to-speech model for voice agents: audio in and audio out over one WebSocket, with built-in barge-in and function calling.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-05","first_response_ms":864,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en"],"status":"preview","benchmarks":["tau3-bench","audio-multichallenge"],"url":"https://smallest.ai/hydra","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/smallest-hydra"}
{"slug":"seeduplex","name":"Seeduplex (Doubao Realtime Voice Model 3.0)","vendor":"ByteDance Seed","tagline":"ByteDance Seed's native full-duplex speech model, which listens while it speaks, runs voice calls in the Doubao app and is sold on Volcengine as Doubao Realtime Voice Model 3.0.","category":"assistant-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-04","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://seed.bytedance.com/en/blog/introducing-seed-full-duplex-speech-llm-attentive-listening-robust-interference-suppression-enabling-more-natural-interaction","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/seeduplex"}
{"slug":"elevenlabs-agents","name":"ElevenLabs Agents (ElevenAgents)","vendor":"ElevenLabs","tagline":"The TTS incumbent's turnkey agent platform — turn-taking, RAG, and telephony pre-wired.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-05","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["90+"],"status":"current","benchmarks":[],"url":"https://elevenlabs.io/conversational-ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/elevenlabs-agents"}
{"slug":"vapi","name":"Vapi","vendor":"Vapi","tagline":"Developer-first orchestration layer for inbound/outbound voice agents with BYO models.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2023-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","es","+100"],"status":"current","benchmarks":[],"url":"https://vapi.ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/vapi"}
{"slug":"retell-ai","name":"Retell AI","vendor":"Retell","tagline":"Low-latency voice agent API built for call-centre workloads and SIP carriers.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-02","first_response_ms":600,"open_weights":false,"license":null,"commercial_use":"yes","languages":["55+"],"status":"current","benchmarks":[],"url":"https://retellai.com","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/retell-ai"}
{"slug":"deepgram-voice-agent","name":"Deepgram Voice Agent","vendor":"Deepgram","tagline":"STT+TTS incumbent's agent API using Nova-3 / Flux for transcription and Aura-2 for voice.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-12","first_response_ms":425,"open_weights":false,"license":null,"commercial_use":"yes","languages":["60+"],"status":"current","benchmarks":[],"url":"https://deepgram.com/product/voice-agents","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/deepgram-voice-agent"}
{"slug":"bland","name":"Bland","vendor":"Bland AI","tagline":"Turnkey voice-agent platform for inbound and outbound telephony with web agents and an API.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2023-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["40+"],"status":"current","benchmarks":[],"url":"https://www.bland.ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/bland"}
{"slug":"polyai-dialog-rsn-1","name":"PolyAI Dialog-RSN-1","vendor":"PolyAI","tagline":"Audio-in, text-out dialog model that folds turn-taking, speech recognition, reasoning, and tool calls into one model and leaves speech output to a separate TTS.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-07","first_response_ms":300,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://poly.ai/blog/PolyAI-dialog-rsn-1","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/polyai-dialog-rsn-1"}
{"slug":"assemblyai-voice-agent-api","name":"AssemblyAI Voice Agent API","vendor":"AssemblyAI","tagline":"AssemblyAI's own speech understanding, LLM reasoning, and voice generation behind a single WebSocket; AssemblyAI is retiring it.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"legacy","benchmarks":[],"url":"https://www.assemblyai.com/docs/voice-agents/voice-agent-api","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/assemblyai-voice-agent-api"}
{"slug":"mistral-voice-agents","name":"Mistral Voice Agents (Voxtral 3)","vendor":"Mistral AI","tagline":"Mistral's voice agent SDK and API: Voxtral realtime transcription with end-of-turn detection, a Mistral model, and Voxtral TTS in one streaming pipeline.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://docs.mistral.ai/studio/audio/voice-agents/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/mistral-voice-agents"}
{"slug":"cartesia-managed-agents","name":"Cartesia Managed Agents","vendor":"Cartesia","tagline":"Cartesia's hosted voice-agent platform, with listening, turn-taking, tool calls and speech tuned together around the LLM of your choice.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://cartesia.ai/blog/managed-agents","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/cartesia-managed-agents"}
{"slug":"agora-conversational-ai-engine","name":"Agora Conversational AI Engine","vendor":"Agora","tagline":"Hosted engine that joins a voice agent to an Agora RTC channel, running a cascaded ASR, LLM and TTS pipeline or a realtime multimodal model with voice interruption.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://www.agora.io/en/products/conversational-ai-engine/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/agora-conversational-ai-engine"}
{"slug":"azure-voice-live","name":"Azure Voice Live API","vendor":"Microsoft","tagline":"Managed Azure API that puts speech recognition, a generative model of your choice and text to speech behind one realtime interface for voice agents.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-05","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["140+"],"status":"current","benchmarks":[],"url":"https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/azure-voice-live"}
{"slug":"fish-agents","name":"Fish Agents","vendor":"Fish Audio","tagline":"Fish Audio's hosted voice-agent platform, in public beta, with a cascade mode and a full-duplex mode built on its Live 3 speech model.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","+49"],"status":"preview","benchmarks":[],"url":"https://docs.fish.audio/agents/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/fish-agents"}
{"slug":"inworld-realtime-api","name":"Inworld Realtime API","vendor":"Inworld AI","tagline":"Realtime session API that connects speech to text, an LLM you choose and text to speech over one WebSocket or WebRTC connection, following the OpenAI Realtime protocol.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://docs.inworld.ai/realtime/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/inworld-realtime-api"}
{"slug":"sarvam-voice-agents","name":"Sarvam Voice Agents (Samvaad)","vendor":"Sarvam AI","tagline":"Sarvam's hosted platform for phone and web voice agents in 11 Indian languages and English, running on speech recognition, LLM and speech synthesis models that Sarvam hosts itself.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["hi","ta","te","bn","mr","gu","kn","ml","pa","or","as","en"],"status":"current","benchmarks":[],"url":"https://docs.sarvam.ai/voice-agents/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/sarvam-voice-agents"}
{"slug":"sierra-voice","name":"Sierra Voice","vendor":"Sierra","tagline":"Voice channel of Sierra's enterprise customer-service agent platform, handling inbound and outbound phone calls with interruption handling in over 55 languages.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["55+"],"status":"current","benchmarks":[],"url":"https://sierra.ai/product/voice","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/sierra-voice"}
{"slug":"smallest-atoms","name":"Smallest AI Voice Agents (Atoms)","vendor":"Smallest AI","tagline":"Smallest AI's hosted platform for phone and web voice agents, built on its Pulse speech to text and Lightning text to speech, with a dashboard, API and Python SDK.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-04","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","hi","+15"],"status":"current","benchmarks":[],"url":"https://smallest.ai/voice-agents","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/smallest-atoms"}
{"slug":"synthflow","name":"Synthflow","vendor":"Synthflow","tagline":"Hosted platform for building, testing and running phone voice agents, with its own telephony plus a choice of speech recognition, LLM and voice models.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2023-12","first_response_ms":500,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es","de","fr","ja","+27"],"status":"current","benchmarks":[],"url":"https://synthflow.ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/synthflow"}
{"slug":"telnyx-ai-assistants","name":"Telnyx AI Assistants","vendor":"Telnyx","tagline":"Telnyx's managed voice agent product: assistants set up in a portal or API that chain speech recognition, an LLM and speech synthesis on Telnyx's network.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es","fr","de","ja","+5"],"status":"current","benchmarks":[],"url":"https://developers.telnyx.com/docs/inference/ai-assistants/no-code-voice-assistant","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/telnyx-ai-assistants"}
{"slug":"tencent-rtc-conversational-ai","name":"Tencent RTC Conversational AI","vendor":"Tencent","tagline":"Voice agent service on Tencent's TRTC real-time network, with built-in speech recognition and interruption handling and the customer's choice of LLM and TTS.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["zh","en","yue","ja","ko","+23"],"status":"current","benchmarks":[],"url":"https://trtc.io/solutions/conversational-ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/tencent-rtc-conversational-ai"}
{"slug":"twilio-conversationrelay","name":"Twilio Conversation Relay","vendor":"Twilio","tagline":"Twilio Voice service that runs speech recognition, speech synthesis and interruption handling on a call and exchanges text with the customer's own LLM application over a WebSocket.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es","fr","de","ja","+24"],"status":"current","benchmarks":[],"url":"https://www.twilio.com/docs/voice/conversationrelay","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/twilio-conversationrelay"}
{"slug":"vogent","name":"Vogent","vendor":"Vogent","tagline":"Hosted platform for building, testing and deploying phone voice agents, with a flow builder, IVR navigation models and Vogent's own conversational LLMs and voices.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","es"],"status":"current","benchmarks":[],"url":"https://www.vogent.ai","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/vogent"}
{"slug":"zego-ai-agent","name":"ZEGO Real-time Interactive AI Agent","vendor":"ZEGO","tagline":"ZEGO's server API and SDKs for voice and digital human calls with an agent, chaining third-party ASR, LLM and TTS over its real-time network.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2024-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["zh","yue","en","ja","ko","+12"],"status":"current","benchmarks":[],"url":"https://www.zegocloud.com/docs/aiagent-server/introduction/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/zego-ai-agent"}
{"slug":"zoom-voice-agent-api","name":"Zoom Voice Agent API","vendor":"Zoom","tagline":"Zoom AI Services API for spoken conversation over one WebSocket, combining Zoom's speech recognition, a built-in or customer-supplied LLM and speech synthesis.","category":"voice-agent-platform","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://developers.zoom.us/docs/ai-services/voice-agent/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/zoom-voice-agent-api"}
{"slug":"livekit-agents","name":"LiveKit Agents","vendor":"LiveKit","tagline":"Open agent framework on top of the LiveKit WebRTC mesh — 50+ provider plugins, first-class FD.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-09","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/livekit/agents","paper_url":null,"repo_url":"https://github.com/livekit/agents","page_url":"https://www.fullduplex.ai/models/livekit-agents"}
{"slug":"pipecat","name":"Pipecat","vendor":"Daily","tagline":"Python framework for voice agents with swappable ASR / LLM / TTS and sane defaults for turn-taking.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2023-12","first_response_ms":null,"open_weights":false,"license":"BSD-2-Clause","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://pipecat.ai","paper_url":null,"repo_url":"https://github.com/pipecat-ai/pipecat","page_url":"https://www.fullduplex.ai/models/pipecat"}
{"slug":"kyutai-unmute","name":"Kyutai Unmute","vendor":"Kyutai","tagline":"Open reference stack that gives any text LLM ears and a voice via delayed streams modelling.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-07","first_response_ms":null,"open_weights":true,"license":"MIT (code) / CC-BY-4.0 (weights)","commercial_use":"yes","languages":["en","fr"],"status":"current","benchmarks":[],"url":"https://github.com/kyutai-labs/unmute","paper_url":null,"repo_url":"https://github.com/kyutai-labs/unmute","page_url":"https://www.fullduplex.ai/models/kyutai-unmute"}
{"slug":"hf-speech-to-speech","name":"HF speech-to-speech","vendor":"Hugging Face","tagline":"Modular reference pipeline that cascades VAD → ASR → LLM → TTS for local voice agents.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-08","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","fr","es","zh","ja","ko"],"status":"current","benchmarks":[],"url":"https://github.com/huggingface/speech-to-speech","paper_url":null,"repo_url":"https://github.com/huggingface/speech-to-speech","page_url":"https://www.fullduplex.ai/models/hf-speech-to-speech"}
{"slug":"qwen-audio-agent","name":"Qwen Audio Agent","vendor":"QwenAudio","tagline":"Apache-2.0 realtime voice runtime that keeps a conversation going while a backend agent runs tools and longer tasks.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-07","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/QwenAudio/qwen-audio-agent","paper_url":"https://arxiv.org/abs/2609.25195","repo_url":"https://github.com/QwenAudio/qwen-audio-agent","page_url":"https://www.fullduplex.ai/models/qwen-audio-agent"}
{"slug":"ava-core","name":"AVA Core (AI Voice Agent for Asterisk)","vendor":"hkjarral","tagline":"Open-source voice agent for Asterisk and FreePBX that runs each call through a realtime voice provider or a modular STT, LLM and TTS pipeline.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-09","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/hkjarral/AVA-AI-Voice-Agent-for-Asterisk","paper_url":null,"repo_url":"https://github.com/hkjarral/AVA-AI-Voice-Agent-for-Asterisk","page_url":"https://www.fullduplex.ai/models/ava-core"}
{"slug":"bolna","name":"Bolna","vendor":"Bolna","tagline":"Open-source Python framework that orchestrates telephone voice agents from ASR, LLM and TTS providers over WebSockets, and the base of Bolna's hosted platform.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-01","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/bolna-ai/bolna","paper_url":null,"repo_url":"https://github.com/bolna-ai/bolna","page_url":"https://www.fullduplex.ai/models/bolna"}
{"slug":"cleans2s","name":"CleanS2S","vendor":"OpenDILab","tagline":"Single-file cascaded speech-to-speech agent prototype with streaming ASR, LLM and TTS over WebSockets, user interruption and proactive response strategies.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-09","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"unclear","languages":["zh"],"status":"current","benchmarks":[],"url":"https://github.com/opendilab/CleanS2S","paper_url":"https://arxiv.org/abs/2506.01268","repo_url":"https://github.com/opendilab/CleanS2S","page_url":"https://www.fullduplex.ai/models/cleans2s"}
{"slug":"cyberverse","name":"CyberVerse","vendor":"Lynpoint","tagline":"Self-hosted framework for real-time voice agents over WebRTC, with persona memory, tools, RAG and an optional digital-human video avatar.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":false,"license":"GPL-3.0","commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/Lynpoint/CyberVerse","paper_url":null,"repo_url":"https://github.com/Lynpoint/CyberVerse","page_url":"https://www.fullduplex.ai/models/cyberverse"}
{"slug":"dograh","name":"Dograh","vendor":"Dograh","tagline":"Open-source, self-hostable voice agent platform with a visual workflow builder, telephony integrations and a Pipecat-based voice pipeline.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-09","first_response_ms":null,"open_weights":false,"license":"BSD-2-Clause","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/dograh-hq/dograh","paper_url":null,"repo_url":"https://github.com/dograh-hq/dograh","page_url":"https://www.fullduplex.ai/models/dograh"}
{"slug":"duplexcascade","name":"DuplexCascade","vendor":"SB Intuitions","tagline":"VAD-free cascaded ASR, LLM and TTS pipeline in which a fine-tuned text LLM decides when to wait, respond, backchannel or stop.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-03","first_response_ms":1724,"open_weights":true,"license":"MIT","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":["fdb-v1","voicebench"],"url":"https://github.com/sbintuitions/DuplexCascade","paper_url":"https://arxiv.org/abs/2603.09180","repo_url":"https://github.com/sbintuitions/DuplexCascade","page_url":"https://www.fullduplex.ai/models/duplexcascade"}
{"slug":"fastrtc","name":"FastRTC","vendor":"Gradio","tagline":"Python library that turns a function into a real-time audio or video stream over WebRTC or WebSockets, with built-in pause detection and turn-taking.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-09","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/gradio-app/fastrtc","paper_url":null,"repo_url":"https://github.com/gradio-app/fastrtc","page_url":"https://www.fullduplex.ai/models/fastrtc"}
{"slug":"fireredchat","name":"FireRedChat","vendor":"FireRedTeam (Xiaohongshu)","tagline":"Self-hosted full-duplex voice agent system built on a LiveKit Agents fork, with its own personalized VAD, end-of-turn detector, ASR and TTS services.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-09","first_response_ms":2341,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/FireRedTeam/FireRedChat","paper_url":"https://arxiv.org/abs/2509.06502","repo_url":"https://github.com/FireRedTeam/FireRedChat","page_url":"https://www.fullduplex.ai/models/fireredchat"}
{"slug":"openai-agents-sdk","name":"OpenAI Agents SDK (voice and realtime agents)","vendor":"OpenAI","tagline":"OpenAI's open agent framework, with a voice pipeline that chains speech-to-text, an agent workflow and text-to-speech, and realtime agents on the Realtime API.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-03","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://openai.github.io/openai-agents-python/","paper_url":null,"repo_url":"https://github.com/openai/openai-agents-python","page_url":"https://www.fullduplex.ai/models/openai-agents-sdk"}
{"slug":"patter","name":"Patter","vendor":"PatterAI","tagline":"Open SDK in Python and TypeScript that puts a voice agent on a phone number, with a choice of provider for each layer of the call.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://getpatter.com","paper_url":null,"repo_url":"https://github.com/PatterAI/Patter","page_url":"https://www.fullduplex.ai/models/patter"}
{"slug":"rapida","name":"Rapida","vendor":"Rapida","tagline":"Self-hostable voice agent orchestration platform written in Go, with pluggable STT, LLM, TTS, VAD and end-of-speech detection plus telephony channels.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-12","first_response_ms":null,"open_weights":false,"license":"GPL-2.0 with Rapida's additional terms","commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://rapida.ai","paper_url":null,"repo_url":"https://github.com/rapidaai/voice-ai","page_url":"https://www.fullduplex.ai/models/rapida"}
{"slug":"realtimevoicechat","name":"RealtimeVoiceChat","vendor":"Kolja Beigel","tagline":"Browser-to-Python voice chat stack that chains RealtimeSTT, a local or OpenAI LLM and RealtimeTTS over WebSockets, with interruption handling.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-04","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/KoljaB/RealtimeVoiceChat","paper_url":null,"repo_url":"https://github.com/KoljaB/RealtimeVoiceChat","page_url":"https://www.fullduplex.ai/models/realtimevoicechat"}
{"slug":"ten-framework","name":"TEN Framework","vendor":"TEN Framework (Agora)","tagline":"Open framework from Agora for real-time voice agents built as graphs of STT, LLM, TTS and speech-to-speech extensions, with interruption passed along as a flush command.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2024-06","first_response_ms":null,"open_weights":false,"license":"Apache-2.0 with Agora's additional conditions","commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://theten.ai","paper_url":null,"repo_url":"https://github.com/TEN-framework/ten-framework","page_url":"https://www.fullduplex.ai/models/ten-framework"}
{"slug":"vision-agents","name":"Vision Agents","vendor":"Stream","tagline":"Stream's open Python framework for voice and video agents: realtime models or STT, LLM and TTS pipelines, with video processors and automatic interruption handling.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-10","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://visionagents.ai","paper_url":null,"repo_url":"https://github.com/GetStream/Vision-Agents","page_url":"https://www.fullduplex.ai/models/vision-agents"}
{"slug":"vocode","name":"Vocode","vendor":"Vocode","tagline":"MIT-licensed Python library for streaming voice conversations with LLMs over phone calls, system audio and Zoom, with swappable transcribers, agents and synthesizers.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2023-02","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://github.com/vocodedev/vocode-core","paper_url":null,"repo_url":"https://github.com/vocodedev/vocode-core","page_url":"https://www.fullduplex.ai/models/vocode"}
{"slug":"x-talk","name":"X-Talk","vendor":"Shanghai Jiao Tong University (X-LANCE Lab)","tagline":"Open cascaded spoken dialogue framework with a pure-Python backend that pauses playback on client-side VAD and confirms the barge-in on the server.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-12","first_response_ms":null,"open_weights":false,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh"],"status":"preview","benchmarks":[],"url":"https://github.com/xcc-zach/xtalk","paper_url":"https://arxiv.org/abs/2512.18706","repo_url":"https://github.com/xcc-zach/xtalk","page_url":"https://www.fullduplex.ai/models/x-talk"}
{"slug":"xiaozhi-esp32-server","name":"xiaozhi-esp32-server","vendor":"xinnan-tech","tagline":"Self-hosted backend for XiaoZhi ESP32 voice devices that runs VAD, ASR, an LLM and TTS on the server and streams speech back to the device.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2025-02","first_response_ms":null,"open_weights":false,"license":"MIT","commercial_use":"yes","languages":["zh","yue","en","ja","ko"],"status":"current","benchmarks":[],"url":"https://github.com/xinnan-tech/xiaozhi-esp32-server","paper_url":null,"repo_url":"https://github.com/xinnan-tech/xiaozhi-esp32-server","page_url":"https://www.fullduplex.ai/models/xiaozhi-esp32-server"}
{"slug":"duplexagent-rsi","name":"DuplexAgent-RSI","vendor":"The Chinese University of Hong Kong, Shenzhen / Amphion Technology","tagline":"Research harness that pairs a full-duplex interaction model with background LLMs and coding agents, and revises its own six modules from failure traces.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["fdb-v1","fdb-v3","voicebench"],"url":"https://amphionteam.github.io/DuplexAgent-RSI-Demo/","paper_url":"https://arxiv.org/abs/2610.11299","repo_url":null,"page_url":"https://www.fullduplex.ai/models/duplexagent-rsi"}
{"slug":"jarvis","name":"Jarvis","vendor":"The Chinese University of Hong Kong, Shenzhen / Tencent Hunyuan / Amphion Technology","tagline":"Open coordination harness that tracks versioned background tasks while a full-duplex speech model keeps talking with the user.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (harness code and adapter weights) / GPL-3.0-or-later (VoxeLibre bridge) / LGPL-2.1-or-later (Luanti patches)","commercial_use":"yes","languages":["en","zh"],"status":"current","benchmarks":["tau3-bench","fdb-v3","fdb-v2"],"url":"https://github.com/JensenYX/Jarvis","paper_url":"https://arxiv.org/abs/2610.23804","repo_url":"https://github.com/JensenYX/Jarvis","page_url":"https://www.fullduplex.ai/models/jarvis"}
{"slug":"lts-voiceagent","name":"LTS-VoiceAgent","vendor":"Meituan / University of Chinese Academy of Sciences","tagline":"Listen-Think-Speak framework for cascaded voice agents that starts reasoning on meaningful transcript prefixes while the user is still talking.","category":"agent-framework","duplex_type":"cascade","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2601.19952","paper_url":"https://arxiv.org/abs/2601.19952","repo_url":null,"page_url":"https://www.fullduplex.ai/models/lts-voiceagent"}
{"slug":"qwen2-5-omni","name":"Qwen2.5-Omni","vendor":"Alibaba Cloud","tagline":"End-to-end multimodal Qwen (text / image / audio / video in, text + speech out) with streaming response.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-03","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh","ja","ko","+"],"status":"legacy","benchmarks":["air-bench","voicebench"],"url":"https://github.com/QwenLM/Qwen2.5-Omni","paper_url":"https://arxiv.org/abs/2503.20215","repo_url":"https://github.com/QwenLM/Qwen2.5-Omni","page_url":"https://www.fullduplex.ai/models/qwen2-5-omni"}
{"slug":"qwen3-omni","name":"Qwen3-Omni","vendor":"Alibaba Cloud","tagline":"Natively end-to-end omni-LLM (text / image / audio / video in, text + speech out).","category":"audio-llm","duplex_type":"full-duplex","full_duplex":true,"released":"2025-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh","ja","+16"],"status":"current","benchmarks":["big-bench-audio","voicebench","mmar","air-bench"],"url":"https://github.com/QwenLM/Qwen3-Omni","paper_url":"https://arxiv.org/abs/2509.17765","repo_url":"https://github.com/QwenLM/Qwen3-Omni","page_url":"https://www.fullduplex.ai/models/qwen3-omni"}
{"slug":"ultravox","name":"Ultravox","vendor":"Ultravox (Inworld)","tagline":"Lightweight audio adapter that bolts onto open LLMs such as Llama, Qwen and GLM to give them ears without retraining.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-08","first_response_ms":null,"open_weights":true,"license":"MIT","commercial_use":"yes","languages":["en","+41"],"status":"current","benchmarks":[],"url":"https://www.ultravox.ai","paper_url":null,"repo_url":"https://github.com/fixie-ai/ultravox","page_url":"https://www.fullduplex.ai/models/ultravox"}
{"slug":"lfm3-audio","name":"LFM3-Audio-2B / LFM3-Omni-2B","vendor":"Liquid AI","tagline":"Small open audio and omni models that run on laptops and edge devices: spoken questions in, streaming text and speech out.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":true,"license":"LFM Open License v1.0","commercial_use":"unclear","languages":["en","+17"],"status":"current","benchmarks":[],"url":"https://www.liquid.ai/blog/lfm3","paper_url":null,"repo_url":"https://huggingface.co/LiquidAI/LFM3-Audio-2B","page_url":"https://www.fullduplex.ai/models/lfm3-audio"}
{"slug":"apriel-voice","name":"Apriel Voice 27B Preview","vendor":"ServiceNow","tagline":"Open speech-in model for reasoning and tool use: it hears the user and returns text and tool calls, with speech synthesis left to the application.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":true,"license":"MIT","commercial_use":"yes","languages":["en","+"],"status":"preview","benchmarks":["eva-bench","tau3-bench"],"url":"https://huggingface.co/ServiceNow-AI/Apriel-Voice-27B-preview","paper_url":"https://arxiv.org/abs/2610.40738","repo_url":"https://huggingface.co/ServiceNow-AI/Apriel-Voice-27B-preview","page_url":"https://www.fullduplex.ai/models/apriel-voice"}
{"slug":"freeze-omni","name":"Freeze-Omni","vendor":"Tencent (VITA-MLLM)","tagline":"Speech-to-speech dialogue model that attaches a streaming speech encoder and decoder to a frozen Qwen2-7B-Instruct, with interruptions handled by chunk-level state prediction.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-11","first_response_ms":null,"open_weights":true,"license":"Tencent licence, academic, research and education use only (GitHub LICENSE) / apache-2.0 tag with acceptable use policy (Hub)","commercial_use":"research-only","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/VITA-MLLM/Freeze-Omni","paper_url":"https://arxiv.org/abs/2411.00774","repo_url":"https://github.com/VITA-MLLM/Freeze-Omni","page_url":"https://www.fullduplex.ai/models/freeze-omni"}
{"slug":"mini-omni2","name":"Mini-Omni2","vendor":"gpt-omni","tagline":"Small open omni model on Qwen2-0.5B that answers image and speech queries with streaming speech, the successor to Mini-Omni.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-10","first_response_ms":null,"open_weights":true,"license":"MIT","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/gpt-omni/mini-omni2","paper_url":"https://arxiv.org/abs/2410.11190","repo_url":"https://github.com/gpt-omni/mini-omni2","page_url":"https://www.fullduplex.ai/models/mini-omni2"}
{"slug":"sophon-realtime","name":"Sophon-Realtime v1 (Tag / Flash / E2E)","vendor":"Adventists AI","tagline":"Open voice model family: a 32B speech LLM listens, decides turns and writes tagged replies, and a separate speech model voices them over an OpenAI-Realtime-style endpoint.","category":"audio-llm","duplex_type":"cascade","full_duplex":false,"released":"2026-11","first_response_ms":450,"open_weights":true,"license":"Apache-2.0 (thinker is DuplexJev-32B-Para, CC BY-NC 4.0 on its own card)","commercial_use":"unclear","languages":["zh","en","it","de","es","fr","pt"],"status":"current","benchmarks":["voicebench","big-bench-audio","fdb-v3"],"url":"https://github.com/adventists-ai/sophon-realtime","paper_url":null,"repo_url":"https://github.com/adventists-ai/sophon-realtime","page_url":"https://www.fullduplex.ai/models/sophon-realtime"}
{"slug":"vita","name":"VITA / VITA-1.5","vendor":"Tencent (VITA-MLLM)","tagline":"Open omni models that take video, image, text and audio, with VITA-1.5 adding its own speech decoder and about 1.5 s interaction latency by the maintainers' account.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-09","first_response_ms":1500,"open_weights":true,"license":"Tencent licence, academic, research and education use only (GitHub LICENSE) / apache-2.0 tag with acceptable use policy (Hub, VITA-1.0)","commercial_use":"research-only","languages":["en","zh"],"status":"current","benchmarks":["librispeech"],"url":"https://github.com/VITA-MLLM/VITA","paper_url":"https://arxiv.org/abs/2501.01957","repo_url":"https://github.com/VITA-MLLM/VITA","page_url":"https://www.fullduplex.ai/models/vita"}
{"slug":"baichuan-omni-1-5","name":"Baichuan-Omni-1.5","vendor":"Baichuan Inc.","tagline":"Open 7B omni model that takes text, image, video and audio and answers in text and speech, with Chinese and English voice conversation.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 + Baichuan-Omni-1.5 Community License Agreement","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/baichuan-inc/Baichuan-Omni-1.5","paper_url":"https://arxiv.org/abs/2501.15368","repo_url":"https://github.com/baichuan-inc/Baichuan-Omni-1.5","page_url":"https://www.fullduplex.ai/models/baichuan-omni-1-5"}
{"slug":"echox","name":"EchoX","vendor":"FreedomIntelligence (CUHK-Shenzhen)","tagline":"English speech-to-speech LLM in 3B and 8B sizes, trained with an echo objective meant to keep the text model's reasoning when it answers in speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/FreedomIntelligence/EchoX","paper_url":"https://arxiv.org/abs/2509.09174","repo_url":"https://github.com/FreedomIntelligence/EchoX","page_url":"https://www.fullduplex.ai/models/echox"}
{"slug":"emova","name":"EMOVA","vendor":"Huawei Noah's Ark Lab and university partners","tagline":"Omni-modal LLM that sees, hears and speaks, with Chinese and English speech output in 24 styles across two speakers, three pitches and four emotions.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-03","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":["librispeech"],"url":"https://emova-ollm.github.io/","paper_url":"https://arxiv.org/abs/2409.18042","repo_url":"https://github.com/emova-ollm/EMOVA","page_url":"https://www.fullduplex.ai/models/emova"}
{"slug":"flashlabs-chroma","name":"FlashLabs Chroma 1.0","vendor":"FlashLabs","tagline":"Open 4B spoken dialogue model that answers in a voice cloned from a few seconds of reference audio, with streaming text and speech output.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["uro-bench"],"url":"https://github.com/FlashLabs-AI-Corp/FlashLabs-Chroma","paper_url":"https://arxiv.org/abs/2601.11141","repo_url":"https://github.com/FlashLabs-AI-Corp/FlashLabs-Chroma","page_url":"https://www.fullduplex.ai/models/flashlabs-chroma"}
{"slug":"flexislm","name":"FlexiSLM","vendor":"CUHK-Shenzhen (Amphion) and ByteDance","tagline":"Spoken language model whose speech frame rate can be set at inference between 12.5 Hz and 4 Hz, trading quality for speed in one checkpoint.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-08","first_response_ms":null,"open_weights":true,"license":"MIT (code) / Apache-2.0 (weights)","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["voicebench","librispeech"],"url":"https://github.com/AmphionTeam/FlexiSLM","paper_url":"https://arxiv.org/abs/2606.31247","repo_url":"https://github.com/AmphionTeam/FlexiSLM","page_url":"https://www.fullduplex.ai/models/flexislm"}
{"slug":"fun-audio-chat","name":"Fun-Audio-Chat","vendor":"Alibaba Cloud (Tongyi Fun Team)","tagline":"Open 8B audio language model for voice conversation with a 5 Hz LLM backbone, whose full-duplex variant is described in the paper but not released.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-12","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh"],"status":"current","benchmarks":["voicebench","librispeech"],"url":"https://github.com/QwenAudio/Fun-Audio-Chat","paper_url":"https://arxiv.org/abs/2512.20156","repo_url":"https://github.com/QwenAudio/Fun-Audio-Chat","page_url":"https://www.fullduplex.ai/models/fun-audio-chat"}
{"slug":"interactiveomni","name":"InteractiveOmni","vendor":"SenseTime","tagline":"Open omni-modal model in 4B and 8B sizes that takes image, video, audio and text and replies with text and streaming speech across multiple turns.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-10","first_response_ms":null,"open_weights":true,"license":"MIT","commercial_use":"yes","languages":["zh","en"],"status":"current","benchmarks":["voicebench","librispeech"],"url":"https://github.com/OpenSenseNova/InteractiveOmni","paper_url":"https://arxiv.org/abs/2510.13747","repo_url":"https://github.com/OpenSenseNova/InteractiveOmni","page_url":"https://www.fullduplex.ai/models/interactiveomni"}
{"slug":"mimo-audio","name":"MiMo-Audio","vendor":"Xiaomi","tagline":"Open 7B audio language model pretrained on over 100 million hours of audio, with a base checkpoint for few-shot tasks and an instruct checkpoint for spoken dialogue.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code) / MIT (weights)","commercial_use":"yes","languages":["en","zh"],"status":"current","benchmarks":["big-bench-audio","mmar","librispeech","seed-tts-eval"],"url":"https://github.com/XiaomiMiMo/MiMo-Audio","paper_url":"https://arxiv.org/abs/2512.23808","repo_url":"https://github.com/XiaomiMiMo/MiMo-Audio","page_url":"https://www.fullduplex.ai/models/mimo-audio"}
{"slug":"mooer-omni","name":"MooER-omni-v1","vendor":"Moore Threads","tagline":"Speech interaction model from Moore Threads that takes a spoken Mandarin question and answers in speech, one turn at a time.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-10","first_response_ms":null,"open_weights":true,"license":"MIT (code) / Apache-2.0 (weights)","commercial_use":"unclear","languages":["zh"],"status":"current","benchmarks":[],"url":"https://github.com/MooreThreads/MooER","paper_url":null,"repo_url":"https://github.com/MooreThreads/MooER","page_url":"https://www.fullduplex.ai/models/mooer-omni"}
{"slug":"moss-speech","name":"MOSS-Speech","vendor":"OpenMOSS (Fudan)","tagline":"Speech-to-speech LLM that generates spoken replies directly, without writing text first, in Chinese and English.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-10","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code) / no licence stated for weights","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/OpenMOSS/MOSS-Speech","paper_url":"https://arxiv.org/abs/2510.00499","repo_url":"https://github.com/OpenMOSS/MOSS-Speech","page_url":"https://www.fullduplex.ai/models/moss-speech"}
{"slug":"opens2s","name":"OpenS2S","vendor":"CASIA (Institute of Automation, Chinese Academy of Sciences)","tagline":"An open end-to-end speech model for empathetic spoken dialogue in English and Chinese, released with weights, training code and its synthetic training data.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-07","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":["voicebench","uro-bench"],"url":"https://github.com/CASIA-LM/OpenS2S","paper_url":"https://arxiv.org/abs/2507.05177","repo_url":"https://github.com/CASIA-LM/OpenS2S","page_url":"https://www.fullduplex.ai/models/opens2s"}
{"slug":"osum-echat","name":"OSUM-EChat","vendor":"ASLP Lab, Northwestern Polytechnical University","tagline":"An open empathetic spoken dialogue model that reasons about the speaker's age, gender, emotion and sound events before it answers in speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-08","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code and weights as tagged) / Qwen Research License on the base LLM","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":["voicebench"],"url":"https://github.com/ASLP-lab/OSUM","paper_url":"https://arxiv.org/abs/2508.09600","repo_url":"https://github.com/ASLP-lab/OSUM","page_url":"https://www.fullduplex.ai/models/osum-echat"}
{"slug":"slam-omni","name":"SLAM-Omni","vendor":"X-LANCE Lab, Shanghai Jiao Tong University / Microsoft","tagline":"A small timbre-controllable speech-to-speech dialogue model trained in a single stage, with its recipe and checkpoints in the SLAM-LLM toolkit.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":true,"license":"MIT (code) / GPL-3.0, research only (Chinese checkpoint)","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":[],"url":"https://github.com/X-LANCE/SLAM-LLM/tree/main/examples/s2s","paper_url":"https://arxiv.org/abs/2412.15649","repo_url":"https://github.com/X-LANCE/SLAM-LLM","page_url":"https://www.fullduplex.ai/models/slam-omni"}
{"slug":"speechgpt","name":"SpeechGPT","vendor":"Fudan University (OpenMOSS)","tagline":"An early open speech-in, speech-out language model: LLaMA-7B trained on discrete speech units to follow spoken instructions and reply in speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2023-09","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code) / no licence stated (weights, initialised from LLaMA-7B)","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/0nutation/SpeechGPT","paper_url":"https://arxiv.org/abs/2305.11000","repo_url":"https://github.com/0nutation/SpeechGPT","page_url":"https://www.fullduplex.ai/models/speechgpt"}
{"slug":"step-audio-r1-1","name":"Step-Audio R1.1","vendor":"StepFun","tagline":"StepFun's open audio reasoning model: a 33.5B-parameter checkpoint that reasons over the audio itself and answers in text.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://huggingface.co/stepfun-ai/Step-Audio-R1.1","paper_url":"https://arxiv.org/abs/2510.09592","repo_url":"https://github.com/stepfun-ai/Step-Audio-R1","page_url":"https://www.fullduplex.ai/models/step-audio-r1-1"}
{"slug":"usdm","name":"USDM (Unified Spoken Dialog Model)","vendor":"NAVER Cloud / Seoul National University","tagline":"A speech-text model built on Mistral-7B that answers spoken input with speech matched to the input's prosody, without a separate speech recognition or text-to-speech stage.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-11","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code and weights) / speech tokenizer from SeamlessM4T (CC-BY-NC-4.0 models)","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/naver-ai/usdm","paper_url":"https://arxiv.org/abs/2402.05706","repo_url":"https://github.com/naver-ai/usdm","page_url":"https://www.fullduplex.ai/models/usdm"}
{"slug":"vita-audio","name":"VITA-Audio","vendor":"VITA-MLLM (Tencent Youtu Lab / Nanjing University)","tagline":"An end-to-end speech model that produces audio tokens in its first forward pass to shorten the wait for the first audio chunk.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-05","first_response_ms":243,"open_weights":true,"license":"Apache-2.0 (Hub weights) / research-only licence text in the repository","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["seed-tts-eval","librispeech"],"url":"https://github.com/VITA-MLLM/VITA-Audio","paper_url":"https://arxiv.org/abs/2505.03739","repo_url":"https://github.com/VITA-MLLM/VITA-Audio","page_url":"https://www.fullduplex.ai/models/vita-audio"}
{"slug":"vita-qinyu","name":"VITA-QinYu","vendor":"Tencent (QinYu Team / VITA Team)","tagline":"An end-to-end spoken language model built for expressive speech: natural conversation, role-play from a character description and, in a checkpoint not yet released, singing.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"Research-only licence (GitHub) / Apache-style terms with an EU exclusion (Hub weights)","commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["uro-bench"],"url":"https://github.com/VITA-MLLM/VITA-QinYu","paper_url":"https://arxiv.org/abs/2605.06765","repo_url":"https://github.com/VITA-MLLM/VITA-QinYu","page_url":"https://www.fullduplex.ai/models/vita-qinyu"}
{"slug":"vocalnet","name":"VocalNet","vendor":"Shanghai Jiao Tong University / Ant Group","tagline":"A family of open speech LLMs that predict several speech tokens per step to answer faster, released with training code and data.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-04","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":[],"url":"https://github.com/SJTU-OmniAgent/VocalNet","paper_url":"https://arxiv.org/abs/2504.04060","repo_url":"https://github.com/SJTU-OmniAgent/VocalNet","page_url":"https://www.fullduplex.ai/models/vocalnet"}
{"slug":"aero-realtime","name":"Aero Realtime","vendor":"The University of Hong Kong / LMMs-Lab / Tsinghua University","tagline":"Open 4B streaming audio-video model that predicts one text token or a silence token for every 80 ms audio slot, so input and output advance together.","category":"audio-llm","duplex_type":"other","full_duplex":false,"released":"2026-08","first_response_ms":null,"open_weights":true,"license":"MIT (weights and data on the Hub, code as declared in pyproject.toml) / no LICENSE file in the GitHub repository","commercial_use":"yes","languages":["en"],"status":"current","benchmarks":[],"url":"https://kcz358.github.io/aero-realtime/","paper_url":"https://arxiv.org/abs/2608.08469","repo_url":"https://github.com/kcz358/aero-realtime","page_url":"https://www.fullduplex.ai/models/aero-realtime"}
{"slug":"e-chat","name":"E-chat","vendor":"ASLP Lab, Northwestern Polytechnical University / Alibaba Group","tagline":"Emotion-sensitive spoken dialogue model that feeds a speech emotion embedding into an LLM and answers Chinese speech in text.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2023-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh"],"status":"current","benchmarks":[],"url":"https://anonymous-echat.github.io/E-chat/","paper_url":"https://arxiv.org/abs/2401.00475","repo_url":null,"page_url":"https://www.fullduplex.ai/models/e-chat"}
{"slug":"emoomni","name":"EmoOmni","vendor":"ASLP Lab, Northwestern Polytechnical University / University of Surrey","tagline":"Emotional dialogue model on Qwen2.5-Omni-7B that reasons about a speaker's face and voice before answering in expressive speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-02","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://github.com/ASLP-lab/EmoOmni","paper_url":"https://arxiv.org/abs/2602.21900","repo_url":"https://github.com/ASLP-lab/EmoOmni","page_url":"https://www.fullduplex.ai/models/emoomni"}
{"slug":"ex-omni","name":"Ex-Omni","vendor":"Tencent (LIGHTSPEED) / The Chinese University of Hong Kong, Shenzhen","tagline":"Omni model on Qwen3-8B that answers speech or text with text, speech and 52-dimensional facial blendshapes for a 3D talking face.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-06","first_response_ms":null,"open_weights":true,"license":"License Terms of Ex-Omni: academic use only, not intended for use in the European Union (GitHub LICENSE.txt) / no licence field on the Hub","commercial_use":"research-only","languages":[],"status":"current","benchmarks":["voicebench"],"url":"https://github.com/Tencent/Ex-Omni","paper_url":"https://arxiv.org/abs/2602.07106","repo_url":"https://github.com/Tencent/Ex-Omni","page_url":"https://www.fullduplex.ai/models/ex-omni"}
{"slug":"moss-audio","name":"MOSS-Audio","vendor":"OpenMOSS (Fudan) / MOSI.AI / Shanghai Innovation Institute","tagline":"Open audio understanding model for speech, environmental sound and music, in 4B and 8B sizes with Instruct and Thinking variants. Audio in, text out.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (weights, per README and Hub) / no LICENSE file for the GitHub code","commercial_use":"unclear","languages":["en","zh"],"status":"current","benchmarks":["mmar"],"url":"https://github.com/OpenMOSS/MOSS-Audio","paper_url":"https://arxiv.org/abs/2606.01802","repo_url":"https://github.com/OpenMOSS/MOSS-Audio","page_url":"https://www.fullduplex.ai/models/moss-audio"}
{"slug":"nuha-speech","name":"Nuha-Speech","vendor":"Elm Company","tagline":"Arabic speech-LLMs at 3B, 7B and 30B, fine-tuned from Qwen-Omni models on 1.5 million Arabic speech instruction samples.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["ar"],"status":"current","benchmarks":["covost2"],"url":"https://arxiv.org/abs/2609.11892","paper_url":"https://arxiv.org/abs/2609.11892","repo_url":null,"page_url":"https://www.fullduplex.ai/models/nuha-speech"}
{"slug":"es4r","name":"ES4R","vendor":"Northeastern University (China)","tagline":"Research framework for empathetic spoken replies that models the affective context of the user's speech before the speech encoder and voices its text reply with StyleTTS2.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://bean0901.github.io/ES4R-Demo/","paper_url":"https://arxiv.org/abs/2601.16225","repo_url":"https://github.com/Bean0901/ES4R","page_url":"https://www.fullduplex.ai/models/es4r"}
{"slug":"flow-omni","name":"Flow-Omni","vendor":"Microsoft / Tsinghua University","tagline":"Research speech-to-speech model in which a Qwen2-0.5B language model predicts continuous mel-spectrogram frames through a flow-matching head instead of discrete codec tokens.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh"],"status":"current","benchmarks":[],"url":"https://cognitivespeech.github.io/flowomni","paper_url":"https://arxiv.org/abs/2412.04917","repo_url":null,"page_url":"https://www.fullduplex.ai/models/flow-omni"}
{"slug":"goat-slm","name":"GOAT-SLM","vendor":"TeleAI (China Telecom)","tagline":"Research spoken language model on TeleChat2-7B that picks up dialect, age, emotion and non-speech sounds in the user's voice and adapts its spoken reply.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en","yue"],"status":"current","benchmarks":[],"url":"https://tele-ai.github.io/GOAT-SLM.github.io/","paper_url":"https://arxiv.org/abs/2507.18119","repo_url":null,"page_url":"https://www.fullduplex.ai/models/goat-slm"}
{"slug":"intrinsicvoice","name":"IntrinsicVoice","vendor":"Fudan University / Alibaba (Tongyi Speech Lab)","tagline":"Research speech-to-speech model on Qwen2-7B that answers in speech tokens without generating text first, grouping the tokens so speech sequences are about as long as text.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2410.08035","paper_url":"https://arxiv.org/abs/2410.08035","repo_url":null,"page_url":"https://www.fullduplex.ai/models/intrinsicvoice"}
{"slug":"loopslm","name":"LoopSLM","vendor":"Tsinghua University / CUHK-Shenzhen / Tencent Hunyuan","tagline":"Research speech-in, text-out model on the Qwen2.5-Omni-7B Thinker that replaces written chain-of-thought with repeated passes through a block of decoder layers when planning an empathetic reply.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["sd-eval"],"url":"https://arxiv.org/abs/2609.37818","paper_url":"https://arxiv.org/abs/2609.37818","repo_url":null,"page_url":"https://www.fullduplex.ai/models/loopslm"}
{"slug":"prime-speech","name":"PRIME-Speech","vendor":"Microsoft","tagline":"Research framework that adds speech output to a frozen speech-to-text LLM by training only an audio post-decoder on the backbone's intermediate hidden states.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["big-bench-audio","vocalbench","covost2"],"url":"https://arxiv.org/abs/2606.30944","paper_url":"https://arxiv.org/abs/2606.30944","repo_url":null,"page_url":"https://www.fullduplex.ai/models/prime-speech"}
{"slug":"pslm","name":"PSLM","vendor":"rinna","tagline":"Research Japanese spoken dialogue model that generates text and speech tokens in parallel streams to cut response delay, with speech recognition done by an external model.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["ja"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2406.12428","paper_url":"https://arxiv.org/abs/2406.12428","repo_url":null,"page_url":"https://www.fullduplex.ai/models/pslm"}
{"slug":"reempathy","name":"ReEmpathy","vendor":"Nankai University / Meituan (LongCat Interaction Team)","tagline":"Research spoken dialogue model, fine-tuned from GLM-4-Voice, that interleaves its spoken reply with written reflections on how empathetic that reply is.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2601.18281","paper_url":"https://arxiv.org/abs/2601.18281","repo_url":null,"page_url":"https://www.fullduplex.ai/models/reempathy"}
{"slug":"saslm","name":"SASLM","vendor":"CUHK-Shenzhen / Tencent Ethereal Audio Lab","tagline":"A 3B speech language model on the Qwen2.5-Omni-3B Thinker that derives expressive intent from its own hidden states to shape the prosody of its spoken reply.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-05","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (weights as tagged) / Qwen Research License on the base model","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://huggingface.co/wangkevin02/SASLM-3B","paper_url":"https://arxiv.org/abs/2604.11424","repo_url":"https://github.com/wangkevin02/SASLM","page_url":"https://www.fullduplex.ai/models/saslm"}
{"slug":"spectron","name":"Spectron","vendor":"Google Research","tagline":"Research model from 2023 that adapts a pre-trained LLM to take speech in and continue or answer in speech, working directly on spectrograms.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2023-05","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://michelleramanovich.github.io/spectron/spectron","paper_url":"https://arxiv.org/abs/2305.15255","repo_url":null,"page_url":"https://www.fullduplex.ai/models/spectron"}
{"slug":"speechmedassist","name":"SpeechMedAssist","vendor":"Fudan University / Shanghai Innovation Institute","tagline":"Open speech language model for multi-turn medical consultation in Mandarin, built by fine-tuning LLaMA-Omni2-7B on medical text and then on synthesised doctor-patient speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code, Hub tag) / academic research only (model card, base model LLaMA-Omni2)","commercial_use":"research-only","languages":["zh"],"status":"current","benchmarks":[],"url":"https://github.com/SirryChen/SpeechMedAssist","paper_url":"https://arxiv.org/abs/2601.04638","repo_url":"https://github.com/SirryChen/SpeechMedAssist","page_url":"https://www.fullduplex.ai/models/speechmedassist"}
{"slug":"latxa-omni","name":"Latxa-Omni","vendor":"HiTZ / Aholab, University of the Basque Country (UPV/EHU)","tagline":"Open Basque speech-to-speech model that rebuilds the LLaMA-Omni architecture on the Latxa-3.1-8B-Instruct backbone, trained on a synthetic Basque version of VoiceAssistant-400K.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-12","first_response_ms":null,"open_weights":true,"license":"No licence stated (code, weights) / Llama 3.1 Community License on the base model","commercial_use":"unclear","languages":["eu"],"status":"current","benchmarks":[],"url":"https://github.com/ansuehu/Latxa-Omni","paper_url":"https://arxiv.org/abs/2610.36911","repo_url":"https://github.com/ansuehu/Latxa-Omni","page_url":"https://www.fullduplex.ai/models/latxa-omni"}
{"slug":"style-talker","name":"Style-Talker","vendor":"Columbia University / University of Rochester","tagline":"Spoken dialogue system that fine-tunes Qwen-Audio to write the reply text and a speaking style from the user's audio, then voices it with StyleTTS 2.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":true,"license":"MIT (code) / no licence stated for the checkpoints","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://github.com/xi-j/Style-Talker","paper_url":"https://arxiv.org/abs/2408.11849","repo_url":"https://github.com/xi-j/Style-Talker","page_url":"https://www.fullduplex.ai/models/style-talker"}
{"slug":"tico","name":"TiCo","vendor":"MIT / National Taiwan University","tagline":"Spoken dialogue model that follows instructions on how long its spoken reply should last, using time markers it writes into its own text plan.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-03","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2603.22267","paper_url":"https://arxiv.org/abs/2603.22267","repo_url":null,"page_url":"https://www.fullduplex.ai/models/tico"}
{"slug":"vocalnet-m2","name":"VocalNet-M2","vendor":"Shanghai Jiao Tong University / Ant Group","tagline":"Speech LLM that generates eight-codebook speech tokens directly with multi-token prediction, removing the flow-matching stage to shorten the wait for the first audio chunk.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2511.10232","paper_url":"https://arxiv.org/abs/2511.10232","repo_url":null,"page_url":"https://www.fullduplex.ai/models/vocalnet-m2"}
{"slug":"vocalnet-mdm","name":"VocalNet-MDM","vendor":"Shanghai Jiao Tong University / Ant Group","tagline":"Streaming speech LLM whose Talker fills in blocks of speech tokens by masked diffusion, in place of token-by-token autoregressive decoding.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-02","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2602.08607","paper_url":"https://arxiv.org/abs/2602.08607","repo_url":null,"page_url":"https://www.fullduplex.ai/models/vocalnet-mdm"}
{"slug":"joyai-talker","name":"JoyAI-Talker","vendor":"JD.com","tagline":"Research speech dialogue system from JD.com that pairs a Thinker built on the 48.9B mixture-of-experts JoyAI-LLM Flash and an instruction-controlled Talker with a separate 1.7B turn-control decoder.","category":"audio-llm","duplex_type":"cascade","full_duplex":false,"released":"2026-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["fdb-v15","covost2","librispeech"],"url":"https://arxiv.org/abs/2608.01119","paper_url":"https://arxiv.org/abs/2608.01119","repo_url":null,"page_url":"https://www.fullduplex.ai/models/joyai-talker"}
{"slug":"minmo","name":"MinMo","vendor":"Alibaba (Tongyi Lab, FunAudioLLM team)","tagline":"About 8B voice interaction model from Alibaba's Tongyi Lab that aligns a speech encoder and a streaming voice decoder to Qwen2.5-7B and adds a full-duplex predictor.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":["air-bench","covost2","librispeech"],"url":"https://qwenaudio.github.io/minmo/","paper_url":"https://arxiv.org/abs/2501.06282","repo_url":null,"page_url":"https://www.fullduplex.ai/models/minmo"}
{"slug":"moderatorlm","name":"ModeratorLM","vendor":"Amazon AGI / IIT Kharagpur","tagline":"Role-conditioned speech LLM for multi-party conversations that reads streaming audio chunks and decides by itself whether the assistant should take the floor.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2606.13544","paper_url":"https://arxiv.org/abs/2606.13544","repo_url":null,"page_url":"https://www.fullduplex.ai/models/moderatorlm"}
{"slug":"paralingpt","name":"ParalinGPT","vendor":"Amazon Alexa AI / National Taiwan University","tagline":"Spoken dialogue language model that reads text plus speech embeddings and predicts the current sentiment, the reply's sentiment and the reply text in one sequence.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2023-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2312.15316","paper_url":"https://arxiv.org/abs/2312.15316","repo_url":null,"page_url":"https://www.fullduplex.ai/models/paralingpt"}
{"slug":"spoken-llm","name":"Spoken-LLM","vendor":"National Taiwan University","tagline":"Llama 2-Chat with a speech emotion encoder that predicts a reply's speaking style and text from how the user's turn was spoken.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2024-02","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2402.12786","paper_url":"https://arxiv.org/abs/2402.12786","repo_url":null,"page_url":"https://www.fullduplex.ai/models/spoken-llm"}
{"slug":"ax-k2-alm","name":"A.X K2 ALM","vendor":"SK Telecom","tagline":"Korean streaming voice conversation model that adds a speech encoder, VAD heads and a speech decoder to a frozen 20B mixture-of-experts LLM.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["ko"],"status":"current","benchmarks":[],"url":"https://huggingface.co/skt/A.X-K2-ALM","paper_url":"https://huggingface.co/skt/A.X-K2-ALM/resolve/main/assets/A.X-K2-ALM-Technical-Report.pdf","repo_url":null,"page_url":"https://www.fullduplex.ai/models/ax-k2-alm"}
{"slug":"av-dialog","name":"AV-Dialog","vendor":"University of Washington / Meta AI Research","tagline":"Streaming spoken dialogue system that reads the speaker's lips as well as the audio to pick out the target speaker and time its turns in noise.","category":"audio-llm","duplex_type":"cascade","full_duplex":false,"released":"2025-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2511.11124","paper_url":"https://arxiv.org/abs/2511.11124","repo_url":null,"page_url":"https://www.fullduplex.ai/models/av-dialog"}
{"slug":"a2-llm","name":"A²-LLM","vendor":"Beijing University of Posts and Telecommunications / Zhongguancun Institute of Artificial Intelligence / East China Normal University","tagline":"Speech-to-speech model built on Step-Audio 2 mini that also generates 3D facial motion for an avatar while it answers spoken questions.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-02","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2602.04913","paper_url":"https://arxiv.org/abs/2602.04913","repo_url":"https://github.com/Hooyoung-for-AI/A2-LLM","page_url":"https://www.fullduplex.ai/models/a2-llm"}
{"slug":"cocktail-talker","name":"Cocktail-Talker","vendor":"Columbia University / University of Washington / University of Illinois Urbana-Champaign","tagline":"Speech LLM for noisy multi-speaker conversations that decides at each turn whether to respond, keep listening or ignore the speech.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2607.27756","paper_url":"https://arxiv.org/abs/2607.27756","repo_url":"https://github.com/xi-j/Cocktail-Talker","page_url":"https://www.fullduplex.ai/models/cocktail-talker"}
{"slug":"diffuspeech","name":"DiffuSpeech","vendor":"National University of Singapore / Tencent","tagline":"Masked-diffusion speech-text language model that writes a text reasoning trace and the spoken answer together.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["librispeech"],"url":"https://arxiv.org/abs/2601.22889","paper_url":"https://arxiv.org/abs/2601.22889","repo_url":null,"page_url":"https://www.fullduplex.ai/models/diffuspeech"}
{"slug":"drvoice","name":"DrVoice","vendor":"Alibaba Group (Tongyi Fun Team)","tagline":"7B parallel speech-text conversation model whose LLM handles speech at 5 Hz while a refined head restores 25 Hz tokens for synthesis.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["voicebench","big-bench-audio"],"url":"https://arxiv.org/abs/2506.09349","paper_url":"https://arxiv.org/abs/2506.09349","repo_url":null,"page_url":"https://www.fullduplex.ai/models/drvoice"}
{"slug":"dual-reasoner","name":"Dual-Reasoner","vendor":"Xiamen University / Jilin University / Zhejiang University","tagline":"Spoken dialogue model on Step-Audio 2 mini that interleaves short thinking blocks with speech so that reasoning does not hold up the audio.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-07","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":["voicebench","mmar"],"url":"https://aclanthology.org/2026.findings-acl.199/","paper_url":"https://aclanthology.org/2026.findings-acl.199/","repo_url":null,"page_url":"https://www.fullduplex.ai/models/dual-reasoner"}
{"slug":"empathy-omni","name":"Empathy Omni","vendor":"Zhejiang University / LIGHTSPEED","tagline":"Speech LLM on Qwen2.5-7B-Instruct that plans an emotion trajectory alongside its text and uses it to colour the spoken reply.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2025-08","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["voicebench"],"url":"https://arxiv.org/abs/2508.18655","paper_url":"https://arxiv.org/abs/2508.18655","repo_url":"https://github.com/W311411/Empathy-Omni","page_url":"https://www.fullduplex.ai/models/empathy-omni"}
{"slug":"echochat","name":"EchoChat","vendor":"The Chinese University of Hong Kong / Microsoft","tagline":"Empathetic speech model that reasons from acoustic perception to the user's mental state before it replies, trained with stage-aware reinforcement learning.","category":"audio-llm","duplex_type":"half-duplex","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":true,"license":"MIT (weights) / Apache-2.0 (code, audio encoder), Hub tag other","commercial_use":"unclear","languages":["en"],"status":"current","benchmarks":[],"url":"https://huggingface.co/ddwang2000/EchoChat","paper_url":"https://arxiv.org/abs/2610.04826","repo_url":null,"page_url":"https://www.fullduplex.ai/models/echochat"}
{"slug":"cartesia-sonic","name":"Cartesia Sonic-3.6","vendor":"Cartesia","tagline":"State-space TTS with ~90 ms time-to-first-audio and realistic non-verbal sounds.","category":"tts","duplex_type":"other","full_duplex":false,"released":"2026-08","first_response_ms":90,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","+41"],"status":"current","benchmarks":[],"url":"https://cartesia.ai/blog/sonic-3.6/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/cartesia-sonic"}
{"slug":"cosyvoice-2","name":"CosyVoice 2","vendor":"Alibaba Tongyi","tagline":"Open expressive TTS with strong zero-shot voice cloning across Japanese, Chinese, and English.","category":"tts","duplex_type":"other","full_duplex":false,"released":"2024-12","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["zh","en","ja","ko"],"status":"legacy","benchmarks":[],"url":"https://github.com/QwenAudio/CosyVoice","paper_url":"https://arxiv.org/abs/2412.10117","repo_url":"https://github.com/QwenAudio/CosyVoice","page_url":"https://www.fullduplex.ai/models/cosyvoice-2"}
{"slug":"qwen3-tts","name":"Qwen3-TTS","vendor":"Alibaba Cloud","tagline":"Open streaming TTS successor to CosyVoice with voice cloning, voice design, and 10-language coverage.","category":"tts","duplex_type":"other","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["en","zh","ja","ko","es","fr","de","it","ru","pt"],"status":"current","benchmarks":[],"url":"https://github.com/QwenLM/Qwen3-TTS","paper_url":"https://arxiv.org/abs/2601.15621","repo_url":"https://github.com/QwenLM/Qwen3-TTS","page_url":"https://www.fullduplex.ai/models/qwen3-tts"}
{"slug":"moss-tts","name":"MOSS-TTS","vendor":"OpenMOSS (Fudan)","tagline":"Real-time streaming TTS family with voice cloning, multi-speaker dialogue, and sound generation.","category":"tts","duplex_type":"other","full_duplex":false,"released":"2026-02","first_response_ms":null,"open_weights":true,"license":"Apache-2.0","commercial_use":"yes","languages":["zh","en","de","+17"],"status":"current","benchmarks":[],"url":"https://github.com/OpenMOSS/MOSS-TTS","paper_url":"https://arxiv.org/abs/2603.18090","repo_url":"https://github.com/OpenMOSS/MOSS-TTS","page_url":"https://www.fullduplex.ai/models/moss-tts"}
{"slug":"gemini-3-8-flash-tts","name":"Gemini 3.8 Flash TTS / Flash-Lite TTS","vendor":"Google DeepMind","tagline":"Google's GA text-to-speech pair in the Gemini API: Flash TTS for studio-grade expressive voice, Flash-Lite TTS for high-volume and real-time voice-agent cascades.","category":"tts","duplex_type":"other","full_duplex":false,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["100+"],"status":"current","benchmarks":[],"url":"https://aistudio.google.com/docs/models/gemini-3.8-flash-tts","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gemini-3-8-flash-tts"}
{"slug":"seamless-m4t-v2","name":"SeamlessM4T v2","vendor":"Meta AI","tagline":"Direct speech-to-speech translation across 100+ languages, with weights in the open.","category":"s2st","duplex_type":"other","full_duplex":false,"released":"2023-11","first_response_ms":null,"open_weights":true,"license":"CC-BY-NC-4.0 (weights) / MIT (code)","commercial_use":"non-commercial","languages":["100+"],"status":"current","benchmarks":[],"url":"https://ai.meta.com/research/seamless-communication/","paper_url":"https://arxiv.org/abs/2312.05187","repo_url":"https://github.com/facebookresearch/seamless_communication","page_url":"https://www.fullduplex.ai/models/seamless-m4t-v2"}
{"slug":"hibiki","name":"Hibiki","vendor":"Kyutai","tagline":"On-device high-fidelity simultaneous speech-to-speech translation — Moshi-style dual-stream for FR → EN, with voice transfer.","category":"s2st","duplex_type":"full-duplex","full_duplex":true,"released":"2025-02","first_response_ms":null,"open_weights":true,"license":"CC-BY-4.0 (weights) / Apache-2.0 & MIT (code)","commercial_use":"yes","languages":["fr","en"],"status":"current","benchmarks":[],"url":"https://kyutai.org/blog/2025-02-10-hibiki/","paper_url":"https://arxiv.org/abs/2502.03382","repo_url":"https://github.com/kyutai-labs/hibiki","page_url":"https://www.fullduplex.ai/models/hibiki"}
{"slug":"gpt-realtime-translate","name":"GPT-Realtime-Translate","vendor":"OpenAI","tagline":"Closed-API live speech-to-speech translation alongside the Realtime API GA — 70+ input languages into 13 output languages at $0.034 / minute.","category":"s2st","duplex_type":"full-duplex","full_duplex":true,"released":"2026-05","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["en","ja","zh","es","fr","+65"],"status":"current","benchmarks":[],"url":"https://developers.openai.com/api/docs/models/gpt-realtime-translate","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gpt-realtime-translate"}
{"slug":"qwen3-8-livetranslate","name":"Qwen3.8-LiveTranslate","vendor":"Alibaba Cloud","tagline":"Alibaba Cloud's simultaneous interpretation API: 60 languages, 29 of them with speech output, with speaker diarization.","category":"s2st","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["60"],"status":"current","benchmarks":[],"url":"https://www.alibabacloud.com/help/en/model-studio/qwen3-5-livetranslate-flash-realtime","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/qwen3-8-livetranslate"}
{"slug":"eleven-realtime-translation","name":"Eleven Realtime Translation","vendor":"ElevenLabs","tagline":"ElevenLabs' live voice translation: each person speaks their own language and hears the other in theirs.","category":"s2st","duplex_type":"other","full_duplex":false,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["90+"],"status":"preview","benchmarks":[],"url":"https://elevenlabs.io/blog/realtime-translation","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/eleven-realtime-translation"}
{"slug":"audiopalm","name":"AudioPaLM","vendor":"Google","tagline":"2023 Google research model that joins PaLM-2 and AudioLM in one decoder for speech recognition, speech translation and speech-to-speech translation.","category":"s2st","duplex_type":"other","full_duplex":false,"released":"2023-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":["covost2"],"url":"https://google-research.github.io/seanet/audiopalm/examples","paper_url":"https://arxiv.org/abs/2306.12925","repo_url":null,"page_url":"https://www.fullduplex.ai/models/audiopalm"}
{"slug":"translatotron","name":"Translatotron","vendor":"Google","tagline":"Sequence-to-sequence model that translates Spanish speech straight into English speech spectrograms, the first of Google's three Translatotron papers.","category":"s2st","duplex_type":"other","full_duplex":false,"released":"2019-04","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["es","en"],"status":"current","benchmarks":[],"url":"https://google-research.github.io/lingvo-lab/translatotron/","paper_url":"https://arxiv.org/abs/1904.06037","repo_url":null,"page_url":"https://www.fullduplex.ai/models/translatotron"}
{"slug":"tavus-griffin-lite","name":"Griffin-Lite","vendor":"Tavus","tagline":"Research preview of Tavus Griffin: one real-time model that watches and hears the user, decides every sub-second whether to speak, react, or wait, and renders its own voice and 720p face.","category":"video-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-10","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"preview","benchmarks":["videofdb"],"url":"https://www.tavus.io/griffin","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/tavus-griffin-lite"}
{"slug":"gemini-3-8-live-avatar","name":"Gemini 3.8 Live with Live Avatar","vendor":"Google DeepMind","tagline":"Gemini 3.8 Live sessions that return a 24 FPS avatar video stream alongside the model's own speech, generally available in Gemini Enterprise.","category":"video-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["97"],"status":"current","benchmarks":[],"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-with-live-avatar/","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/gemini-3-8-live-avatar"}
{"slug":"tavus-cvi","name":"Tavus CVI (Phoenix-4.5 / Sparrow-2 / Raven-1)","vendor":"Tavus","tagline":"Tavus's production video-agent stack: camera perception, its own turn-taking model, STT, LLM, TTS, and a real-time face that keeps reacting while the user talks.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-09","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":["42+"],"status":"current","benchmarks":["turnbench"],"url":"https://docs.tavus.io/sections/conversational-video-interface/overview-cvi","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/tavus-cvi"}
{"slug":"runway-characters","name":"Runway Characters","vendor":"Runway","tagline":"Real-time video agent API on Runway's GWM-1 world model: one image becomes a conversational character that keeps expressing while it speaks and listens.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-03","first_response_ms":1750,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://runway.com/news/introducing-runway-characters","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/runway-characters"}
{"slug":"heygen-liveavatar","name":"HeyGen LiveAvatar (FULL mode)","vendor":"HeyGen","tagline":"HeyGen's real-time avatar service; in FULL mode HeyGen runs speech detection, the LLM, TTS, and turn management, and switches the avatar into a listening state.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2025-11","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"yes","languages":[],"status":"current","benchmarks":[],"url":"https://www.liveavatar.com/","paper_url":null,"repo_url":"https://github.com/heygen-com/liveavatar-web-sdk","page_url":"https://www.fullduplex.ai/models/heygen-liveavatar"}
{"slug":"avtr-1","name":"AVTR-1","vendor":"Avaturn","tagline":"Open stack for real-time interactive avatars: a 153M-parameter motion model conditioned on both speakers' audio, so the face listens back while an external voice agent runs the turns.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-05","first_response_ms":null,"open_weights":true,"license":"AVTR-1 Community License (weights) / PolyForm Noncommercial 1.0.0 (renderer, streamer)","commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://avtr-1.avaturn.live/","paper_url":"https://arxiv.org/abs/2609.22913","repo_url":"https://github.com/avaturn-live/avtr-1","page_url":"https://www.fullduplex.ai/models/avtr-1"}
{"slug":"avatar-forcing","name":"Avatar Forcing","vendor":"KAIST / NTU Singapore / DeepAuto.ai","tagline":"Research head avatar that takes the user's audio and head motion as live inputs and reacts to speech, nods, and laughter while it listens.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-01","first_response_ms":null,"open_weights":true,"license":"CC BY-NC 4.0","commercial_use":"non-commercial","languages":[],"status":"current","benchmarks":[],"url":"https://taekyungki.github.io/AvatarForcing/","paper_url":"https://arxiv.org/abs/2601.00664","repo_url":"https://github.com/TaekyungKi/AvatarForcing","page_url":"https://www.fullduplex.ai/models/avatar-forcing"}
{"slug":"anam","name":"Anam (Cara-4.1 Express)","vendor":"Anam","tagline":"Hosted avatar personas: in the default Turnkey setup Anam runs STT, turn-taking, the LLM, TTS and the real-time face, and user speech interrupts the persona.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-10","first_response_ms":1240,"open_weights":false,"license":null,"commercial_use":"yes","languages":["70+"],"status":"current","benchmarks":[],"url":"https://anam.ai/docs/introduction/overview","paper_url":null,"repo_url":null,"page_url":"https://www.fullduplex.ai/models/anam"}
{"slug":"duo-v1","name":"Duo v1","vendor":"Ant Group (inclusionAI)","tagline":"Open preview of an audio-visual conversation model: every 0.7 s a language module decides to speak or wait, and a diffusion module renders the character's video and voice.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-11","first_response_ms":null,"open_weights":true,"license":"Apache-2.0 (code) / MiniMax H3 Community License (weights)","commercial_use":"unclear","languages":["zh","en"],"status":"preview","benchmarks":[],"url":"https://duo-v1.github.io/","paper_url":"https://arxiv.org/abs/2611.15131","repo_url":"https://github.com/inclusionAI/Duo","page_url":"https://www.fullduplex.ai/models/duo-v1"}
{"slug":"dyaplex","name":"DyaPlex","vendor":"NVIDIA / HKUST","tagline":"Research model that adds a streaming body-motion pathway to a frozen PersonaPlex, so the agent perceives the partner's speech and motion while generating its own.","category":"video-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://research.nvidia.com/labs/amri/projects/DyaPlex","paper_url":"https://arxiv.org/abs/2606.03874","repo_url":null,"page_url":"https://www.fullduplex.ai/models/dyaplex"}
{"slug":"infp","name":"INFP","vendor":"ByteDance","tagline":"Research head-video generator driven by both speakers' audio: one portrait image becomes an agent that moves between speaking and listening without a manual role switch.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2024-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://grisoon.github.io/INFP/","paper_url":"https://arxiv.org/abs/2412.04037","repo_url":null,"page_url":"https://www.fullduplex.ai/models/infp"}
{"slug":"lpm-1-0","name":"LPM 1.0","vendor":"Ailing Zeng et al.","tagline":"Research video model for conversational characters: a 17B diffusion transformer distilled into a streaming generator that renders listening video from user audio and speaking video from synthesized audio.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-04","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://large-performance-model.github.io","paper_url":"https://arxiv.org/abs/2604.07823","repo_url":null,"page_url":"https://www.fullduplex.ai/models/lpm-1-0"}
{"slug":"mimictalker","name":"MimicTalker","vendor":"Xi'an Jiaotong University / Nanjing University / Ant Group","tagline":"Research model for real-time 3D head motion in two-person conversation: it reads the interlocutor's audio and head motion frame by frame and reacts while speaking or listening.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://nuo1wang.github.io/MimicTalker/","paper_url":"https://openaccess.thecvf.com/content/CVPR2026/html/Wang_MimicTalker_A_Multimodal_Interactive_and_Memory-Enhanced_Framework_for_Real-Time_Dyadic_CVPR_2026_paper.html","repo_url":null,"page_url":"https://www.fullduplex.ai/models/mimictalker"}
{"slug":"mio","name":"Mio (Multimodal Interactive Omni-Avatar)","vendor":"Shanda AI Research Tokyo / The University of Tokyo","tagline":"Research digital-human system in five modules (Thinker, Talker, Face Animator, Body Animator, Renderer), whose face animator produces listening motion from the other speaker's audio.","category":"video-fd","duplex_type":"cascade","full_duplex":false,"released":"2025-12","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["en","zh","ja"],"status":"current","benchmarks":[],"url":"https://alayalab.github.io/project_mio_page/","paper_url":"https://arxiv.org/abs/2512.13674","repo_url":null,"page_url":"https://www.fullduplex.ai/models/mio"}
{"slug":"moshi-face","name":"Moshi-Face","vendor":"Nagoya University","tagline":"Research extension of Moshi that reads the user's audio and 3D face tokens and generates the system's speech and facial motion together at 12.5 Hz.","category":"video-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":[],"status":"current","benchmarks":[],"url":"https://arxiv.org/abs/2606.21970","paper_url":"https://arxiv.org/abs/2606.21970","repo_url":null,"page_url":"https://www.fullduplex.ai/models/moshi-face"}
{"slug":"wan-streamer","name":"Wan-Streamer","vendor":"Alibaba Group (Wan Team)","tagline":"Research model: one causal Transformer takes the user's audio and video and generates the agent's speech and video in 160 ms units, including listening behaviour and interruptions.","category":"video-fd","duplex_type":"full-duplex","full_duplex":true,"released":"2026-06","first_response_ms":null,"open_weights":false,"license":null,"commercial_use":"unclear","languages":["zh","en"],"status":"current","benchmarks":[],"url":"https://wan-streamer.com","paper_url":"https://arxiv.org/abs/2606.25041","repo_url":null,"page_url":"https://www.fullduplex.ai/models/wan-streamer"}