We’ve been experimenting with how far a tiny model can go when it’s good at calling external tools - and have just released Jan-nano, a 4 B model trained for MCP. Jan-nano: - tops DeepSeek-V3-671B on MCP tool-use (SimpleQA 80.7%) - handles live web search and multi-step deep research - runs fully on-device (≈4GB VRAM) Tech notes - Base: Qwen3-4B - Fine-tuning: DAPO - We're going to release the full technical report soon Links - Demo tweet: https://x.com/menloresearch/status/1934809407604576559 - Model + GGUF: https://huggingface.co/collections/Menlo/jan-nano-684f6ebfe9... - Jan Beta desktop (viewer/runner): https://jan.ai/docs/desktop/beta How to try it: - Install Jan Beta (macOS/Win/Linux): https://jan.ai/docs/desktop/beta - Go Jan Hub and download Jan-nano (onboarding steps help you to download it) - Get your free Serper API key to test deep research & real-time web search: https://serper.dev/ - Settings -> MCP -> paste your SERPER_API_KEY (gives the model web search access). We’re testing Jan-nano inside Jan's beta (an open-source ChatGPT alternative). Feedback on both the model and the app is very welcome. If setup feels clunky, follow us on X for a short walkthrough video (coming soon) or join our community chat. - X: https://x.com/menloresearch - Discord: https://discord.gg/Exe46xPMbK Huge credit to the Qwen team for the base model. |