StudioTether AI releases BotPsy-0.8B, an Apache 2.0 on-device tool-calling model that doubles Qwen3.5-0.8B agentic scores
Original title: qvac/BotPsy-0.8B · Hugging Face - On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model
Tether AI Research released BotPsy-0.8B, a 0.8B tool-calling model built on Qwen3.5-0.8B and licensed under Apache 2.0, roughly doubling its base model's agentic scores (BFCL v4 23.4 to 44.0, ACEBench 30.1 to 58.7, τ-bench 23.2 to 51.5). It leads the tested models up to 1.2B parameters on six tool-use benchmarks and beats Qwen3.5-2B-Think on five of six, while remaining best suited to low-to-medium-difficulty tasks; BF16 and GGUF weights are available, and Transformers, vLLM and llama.cpp run it without custom code (enable_thinking=True required).
The card provides full benchmark comparisons against same-size models and deployment settings, helping readers gauge its practical fit for on-device tool calling.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.