Tether’s BotPsy AI agent excels in device-based tasks

Tether’s BotPsy AI agent, with 800M parameters, outperformed much larger models in tool-use benchmarks and excels at complex, multi-step tasks on personal devices, while enhancing user privacy.

Tether has introduced QVAC BotPsy, an AI agent with 800 million parameters, optimized for on-device execution of complex, multi-step tasks such as adjusting phone settings, messaging, and scheduling. Despite its compact size, BotPsy outperformed models up to 2.5 times larger in five out of six tool-use benchmarks, including BFCL v4, ACEBench, τ-bench, Claw-Eval, BrowseComp-Plus, and ToolSandbox. The model nearly doubled the scores of its base architecture, Qwen3.5-0.8B-Think, particularly excelling in function-calling and tool-use abilities. BotPsy operates as an agentic system, capable of identifying necessary tools, executing commands sequentially, and adapting to errors—distinguishing it from traditional chatbots. It can perform tasks locally, reducing the need to send user data to the cloud and thereby enhancing privacy. Additionally, BotPsy achieved a leading hallucination resistance score among models with fewer than 1.2 billion parameters. This release expands Tether's QVAC ecosystem, underscoring the efficiency and specialization of smaller AI models for personal devices.