---
title: THOX.ai
---

# THOX.ai

Edge and on-device language models. We build small models that run on hardware
people actually own — laptops, USB sticks, single-board computers, and
microcontrollers.

**14 models are published here.** Everything listed below is real and
downloadable; nothing on this page links to a repository you cannot open.

---

## Open — no access request

| model | what it is | size |
|---|---|---|
| [**ThoxMicro-1bit-9M**](https://huggingface.co/Thox-ai/ThoxMicro-1bit-9M) | BitNet b1.58 **ternary** LM, trained from scratch. Writes children's stories; not an assistant. Research artifact for microcontroller-class targets. | 6.3 MB |
| [**ThoxMicro-1bit-16M**](https://huggingface.co/Thox-ai/ThoxMicro-1bit-16M) | The same model at 16 layers instead of 8. 10.7% lower perplexity; needs more device headroom. Neither supersedes the other. | 8.0 MB |
| [**thox-micro-125m**](https://huggingface.co/Thox-ai/thox-micro-125m) · [GGUF](https://huggingface.co/Thox-ai/thox-micro-125m-GGUF) | 123.6M base LM trained from scratch. Completion only. | 81–247 MB |
| [**ThoxSpeech-ASR-BitNet**](https://huggingface.co/Thox-ai/ThoxSpeech-ASR-BitNet) | BitNet speech recognition, redistributed byte-identical with published SHA-256. | 0.7–1.0 GB |

## Access-gated

Adapters and derived models. Request access on the model page.

| model | base | type |
|---|---|---|
| [ThoxMini-3B](https://huggingface.co/Thox-ai/ThoxMini-3B) | Llama-3.2-3B-Instruct | LoRA + GGUF |
| [ThoxForge-7B](https://huggingface.co/Thox-ai/ThoxForge-7B) | Mistral-7B-Instruct-v0.3 | LoRA + GGUF |
| [ThoxGlobal-7B](https://huggingface.co/Thox-ai/ThoxGlobal-7B) | Qwen2.5-7B-Instruct | LoRA + GGUF |
| [ThoxWave-8B](https://huggingface.co/Thox-ai/ThoxWave-8B) | dolphin-2.9-llama3-8b | LoRA + GGUF |
| [ThoxNova-12B-Core](https://huggingface.co/Thox-ai/ThoxNova-12B-Core) · [Agent](https://huggingface.co/Thox-ai/ThoxNova-12B-Agent) | Gemma-4-12B-it | LoRA + GGUF |
| [fable5-agentic-qwen2.5-14b-lora](https://huggingface.co/Thox-ai/fable5-agentic-qwen2.5-14b-lora) | Qwen2.5-14B-Instruct | QLoRA adapter |

> **On the adapter bases:** ThoxMini, ThoxForge, ThoxGlobal and both ThoxNova
> adapters were trained against `unsloth/*` re-uploads of the listed models, not
> the original repositories. Each card's `base_model` now states the re-upload it
> was actually trained on, with the upstream lineage recorded in the body.
> ThoxWave-8B trains directly on `dphn/dolphin-2.9-llama3-8b`.

## Documentation-only

These contain **no weights**, deliberately, and say so on their own pages:

- [ThoxForge-7B-GGUF](https://huggingface.co/Thox-ai/ThoxForge-7B-GGUF) — the
  quantized build did not pass our factuality gate; publishing it anyway to make
  the repo look complete would route around that control.
- [ThoxWebby-Gemma-4-E2B](https://huggingface.co/Thox-ai/ThoxWebby-Gemma-4-E2B) —
  the browser app pulls upstream weights; we do not re-host them.

---

## Where our models run

- **Ollama** — [`ollama.com/thox-ai`](https://ollama.com/thox-ai)
- **llama.cpp** — every GGUF above
- **ThoxKey / ThoxKey Pro** — USB sticks that carry a model set and a local runtime
- **ThoxMicro / ThoxOS devices** — ESP32-S3 and Raspberry Pi class hardware

## How we publish

- **Measured, not estimated.** Numbers on our cards come from runs we did. Where
  a figure is inherited from an upstream benchmark, the card says so.
- **Negative results ship too.** ThoxMicro-1bit-9M's card leads with the fact
  that ternary quantization costs it 21.4% perplexity, and that the penalty grew
  rather than shrank with more training data.
- **Empty is better than misleading.** A repo with no weights and an honest
  explanation beats one with weights that failed their gate.
- **Licence honesty.** Where downstream terms are genuinely unsettled we say
  `license: other` and explain, rather than picking a permissive licence because
  it is convenient.

## Contact

[thox.ai](https://thox.ai)