BotPsy-0.8B: Tether's sub-billion on-device tool-calling model
Tether AI Research released BotPsy, a 0.8B-parameter model built on Qwen3.5-0.8B and post-trained for function calling and agentic workflows, with open weights under Apache 2.0 in safetensors and GGUF for on-device use via llama.cpp.
Why it matters
A sub-1B model that roughly doubles its base model's tool-use scores (BFCL v4 23.4→44.0, τ-bench 23.2→51.5) and beats a 2.5x larger model on five of six agentic benchmarks, making local tool routing viable without a cloud call.
What you could build with it
An indie developer building a privacy-first phone assistant could run BotPsy locally to handle calendar updates, file operations, and messaging tool calls on-device, only escalating to a frontier cloud model when the small model declines or the task exceeds its scope.
Does it hold up?
Usable today for experimentation — weights are live on Hugging Face and run under llama.cpp — but all benchmarks are Tether's own, the model still makes mistakes and is explicitly not for high-stakes decisions, and no independent deployment reports exist yet.
Built with BotPsy-0.8B
- BotPsy Tests How Far a 0.8B On-Device Tool Agent Can Gohuggingface.blog · HF's own analysis: treats BotPsy as a candidate for tightly scoped local tool routing and warns to validate each workflow with simulations and explicit action limits.
- Tether BotPsy-0.8B: open on-device tool-calling modelpromptea.me · Independent writeup of the four-stage training recipe (SFT, multi-teacher distillation, two RL rounds) and the licensing wrinkle that the blog frames it for research/educational use.
- qvac/BotPsy-0.8B-GGUFshuggingface · The GGUF weight page: BF16, Q8_0 and Q4_K_M builds, with the 4-bit version needing about 2.9 GB VRAM.
Learn more
First spotted on hf: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.