Tether AI Research has launched BotPsy, an open-weight AI model featuring 0.8 billion parameters. Released on November 11, 2026, the model is designed for agentic tasks such as function calling, document management, and email organization.
In tool-use benchmarks, BotPsy outperformed models up to 2.5 times its size. Across six evaluations, it achieved a mean score of 45.0, surpassing the 2-billion-parameter Qwen3.5-2B-Think model in five of those tests.
The model is built on the Alibaba Qwen3.5-0.8B base. Tether applied fine-tuning to improve its performance, resulting in a hallucination resistance score of 86.1 on the BFCL v4 benchmark, the highest among models under 1.2 billion parameters.
BotPsy is distributed under the Apache 2.0 license and is available in GGUF format for local hardware execution. It is part of the broader QVAC initiative, which aims to develop decentralized AI capable of running on local devices without cloud dependency.
Tether intends the model for research and educational use, specifically for low-to-medium difficulty tasks. The company notes that while BotPsy is efficient for narrow agentic functions, it is not intended to replace larger general-purpose systems.
