Datasets:

open-weights
/

open-weight-model-registry

repo_id string	organization string	model_family string	model_name string	architecture_class string	modalities list	total_parameters_b float64	active_parameters_b float64	context_length_native_tokens int64	context_length_extended_tokens int64	weights_available bool	weight_format list	access_status string	license_id string	license_class string	commercial_use_status string	modification_status string	redistribution_status string	self_hosting_status string	documented_runtimes list	verification_status string	last_verified timestamp[s]	source_urls list	notes string
openai/gpt-oss-120b	openai	gpt-oss	gpt-oss-120b	Mixture-of-Experts decoder-only language model	[ "text" ]	117	5.1	131,072	null	true	[ "safetensors", "MXFP4" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers", "vLLM", "SGLang", "Ollama", "LM Studio" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/openai/gpt-oss-120b" ]	OpenAI describes gpt-oss-120b as an open-weight model with 117B total and 5.1B active parameters. The official card documents Apache-2.0 and MXFP4 deployment.
openai/gpt-oss-20b	openai	gpt-oss	gpt-oss-20b	Mixture-of-Experts decoder-only language model	[ "text" ]	21	3.6	131,072	null	true	[ "safetensors", "MXFP4" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers", "vLLM", "SGLang", "Ollama", "LM Studio" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/openai/gpt-oss-20b" ]	OpenAI describes gpt-oss-20b as an open-weight model with 21B total and 3.6B active parameters. The official card documents Apache-2.0.
Qwen/Qwen3-235B-A22B	Qwen	Qwen3	Qwen3-235B-A22B	Mixture-of-Experts causal language model	[ "text" ]	235	22	32,768	131,072	true	[ "safetensors" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/Qwen/Qwen3-235B-A22B" ]	Qwen documents 235B total and 22B activated parameters. The base Qwen3 release uses Apache-2.0; extended context depends on long-context configuration.
Qwen/Qwen3-32B	Qwen	Qwen3	Qwen3-32B	Dense decoder-only causal language model	[ "text" ]	32.8	32.8	32,768	131,072	true	[ "safetensors" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers", "llama.cpp", "vLLM", "SGLang" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/Qwen/Qwen3-32B" ]	Qwen documents 32.8B parameters, 32,768 native context and validation to 131,072 tokens with YaRN.
Qwen/Qwen3-14B	Qwen	Qwen3	Qwen3-14B	Dense decoder-only causal language model	[ "text" ]	14.8	14.8	32,768	131,072	true	[ "safetensors" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/Qwen/Qwen3-14B" ]	Qwen documents 14.8B parameters and context up to 131,072 tokens with YaRN.
deepseek-ai/DeepSeek-R1	deepseek-ai	DeepSeek-R1	DeepSeek-R1	Mixture-of-Experts reasoning language model	[ "text" ]	671	37	131,072	null	true	[ "safetensors" ]	public	mit	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/deepseek-ai/DeepSeek-R1", "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/main/LICENSE" ]	DeepSeek documents 671B total, 37B activated parameters and 128K context. The repository license is MIT.
zai-org/GLM-4.5	zai-org	GLM-4.5	GLM-4.5	Mixture-of-Experts causal language model	[ "text" ]	355	32	131,072	null	true	[ "safetensors", "BF16" ]	public	mit	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers", "vLLM", "SGLang" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/zai-org/GLM-4.5" ]	Z.ai documents GLM-4.5 as 355B-A32B and states that the released model is under MIT and can be used commercially and for secondary development.
mistralai/Mistral-Small-3.1-24B-Instruct-2503	mistralai	Mistral Small 3.1	Mistral-Small-3.1-24B-Instruct-2503	Dense multimodal language-and-vision model	[ "text", "image" ]	24	24	131,072	null	true	[ "safetensors" ]	public	apache-2.0	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "vLLM", "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503" ]	Mistral documents 24B parameters, text+vision input, 128K context and Apache-2.0.
meta-llama/Llama-4-Scout-17B-16E-Instruct	meta-llama	Llama 4	Llama-4-Scout-17B-16E-Instruct	Mixture-of-Experts multimodal language-and-vision model	[ "text", "image" ]	109	17	10,000,000	null	true	[ "safetensors" ]	gated	llama-4-community-license	custom-community-license	subject_to_custom_license_terms	subject_to_custom_license_terms	subject_to_custom_license_terms	supported_after_access	[ "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct" ]	Meta documents Scout as 109B total / 17B activated, natively multimodal, with a 10M-token context and a custom Llama 4 Community License. Access is gated on Hugging Face.
google/gemma-3-27b-it	google	Gemma 3	gemma-3-27b-it	Dense multimodal language-and-vision model	[ "text", "image" ]	27	27	131,072	null	true	[ "safetensors" ]	gated	gemma	custom-terms	subject_to_gemma_terms	subject_to_gemma_terms	subject_to_gemma_terms	supported_after_accepting_terms	[ "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/google/gemma-3-27b-it", "https://ai.google.dev/gemma/terms" ]	Gemma 3 27B is a multimodal open-weight model with 128K context. Hugging Face access requires acknowledgement of the Gemma usage terms.
microsoft/phi-4	microsoft	Phi-4	phi-4	Dense decoder-only Transformer	[ "text" ]	14	14	16,384	null	true	[ "safetensors" ]	public	mit	permissive-open-source	permitted_by_license	permitted_by_license	permitted_by_license	supported	[ "Transformers" ]	verified	2026-09-24T00:00:00	[ "https://huggingface.co/microsoft/phi-4", "https://huggingface.co/microsoft/phi-4/blob/main/LICENSE" ]	Microsoft documents Phi-4 as a 14B dense text model with a 16K context window under the MIT license.

Open-Weight Model Registry

Interactive registry and direct comparison for open-weight AI models

Open-Weight Model Registry is a data-driven Hugging Face Space for exploring and directly comparing downloadable AI models across:

license
access status
architecture
parameter count
context length
modalities
self-hosting
commercial-use status
modification and redistribution status
documented runtimes
verification date

The Space reads from:

open-weights/open-weight-model-registry

Open weights are not the same as open source.

The registry intentionally separates weight availability from license and usage conditions.

Compact Model Detail View

The registry includes a compact detail view for each model.

Selecting Details opens a focused summary with:

repository and publisher
architecture
modalities
total and active parameters
native and extended context
access status
license and license class
commercial-use status
modification and redistribution status
self-hosting status
weight formats
documented runtimes
verification date
notes
primary source links

From the detail view, a model can be sent directly into either side of the comparison view.

Direct Model Comparison

The Space includes a dedicated comparison view for two selected models.

It compares, side by side:

repository and organization
architecture
modalities
total parameters
active parameters
native context length
extended context length
access status
license
license class
commercial-use status
modification status
redistribution status
self-hosting status
weight formats
documented runtimes
last verified date
primary source links

Differences are highlighted to make the underlying metadata easier to inspect.

The interface does not rank models and does not select a winner.

Registry Filters

The registry view supports filtering by:

text search
license
access status
license class
architecture
modality
minimum context length
maximum parameter count

Data Source

Primary dataset:

open-weights/open-weight-model-registry

The Space loads the current train.jsonl directly from the dataset repository.

A small embedded fallback snapshot is included so the interface remains usable if remote loading is temporarily unavailable.

Important License Note

The registry is not legal advice.

Always review the current:

model card
license
access conditions
downstream-use requirements
redistribution terms

before deploying or redistributing a model.

Collaboration & Partnerships

The Open Weights organization is open to collaboration with model developers, inference providers, hardware companies, AI infrastructure teams, model-serving platforms, research groups and open-source projects.

Relevant areas include: