// SUPPORTED MODELS
Local (ONNX Runtime) — profiles exist; routing is experimental
These families have local profiles and loader support in the source. That is a configuration surface, not a conformance claim: local bootstrap, selection, and provider parity are unfinished, so a configured local profile does not yet guarantee your query ran locally. Artifacts are large and may require Hugging Face access.
| FAMILY | SIZES | FORMAT | NOTES |
|---|---|---|---|
| Qwen 2.5 | 1.5B · 3B · 7B · 14B | ONNX | Default local family · size chosen by detected RAM |
| Llama 3 | 1B · 3B · 8B | ONNX | Meta · general purpose |
| Gemma 2/3 | 1B · 2B · 9B | ONNX | Google · strong reasoning |
| Mistral | 7B | ONNX | Mistral model family · third-party ONNX exports from the microsoft and nvidia organizations on Hugging Face; no onnx-community export |
| Phi-3/4 | 3.8B · 14B | ONNX | Microsoft · small + capable |
| DeepSeek Coder | 1.3B · 6.7B | ONNX | Optimised for code generation |
CoreML execution provider on Apple Silicon · CUDA/ROCm/CPU on Linux via ONNX Runtime · Candle is an alternative backend · see MODEL_BACKEND_STATUS.md, a dated backend investigation rather than routing evidence
Cloud providers — profiles offered by finch setup
| PROVIDER | AUTHENTICATION | NOTES |
|---|---|---|
| Claude (Anthropic) | API key | Enter an explicit model ID, or refresh the authenticated catalogue |
| OpenAI Platform | API key | A model ID your Platform account can reach on Chat Completions |
| ChatGPT subscription | Device login — finch auth login | Finch-owned transport, pinned to a recorded Codex compatibility commit; drift fails closed. Experimental. Subscription credentials are not API keys and never cross-bind with Platform ones |
| Grok (xAI) | API key | API billing is separate from Grok/X subscriptions · console.x.ai |
| Gemini (Google) | API key | AI Studio transport; custom endpoints and paths are rejected |
| Mistral | API key | European provider with open-weight model families |
| Groq | API key | Runs open models on the provider's own hardware |
| Ollama · remote Finch daemon | Profile types in source | Configurable; not covered by conformance evidence |
Switch between configured profiles with /model <name> (/provider and /teacher remain compatibility aliases) · Conversation history is preserved across a switch · Streaming and tool support vary by provider and model, and cross-provider conformance coverage is still being published · Model availability changes independently of Finch, so pricing and model rankings are deliberately not listed here