Tether announced the release of BotPsy, an 800‑million‑parameter AI agent designed to perform multi‑step tasks on personal devices such as adjusting phone settings, sending messages and managing schedules. In the company’s internal testing, BotPsy outperformed larger models on five of six tool‑use benchmarks.
Benchmark performance
On the BFCL v4 benchmark, which measures an AI system’s ability to select and execute software functions, BotPsy scored 44.0%. The underlying Qwen3.5‑0.8B model scored 23.4% and a larger 2‑billion‑parameter version of the same architecture scored 37.7%.
On the ACEBench benchmark, which evaluates the accuracy of function calls, BotPsy achieved 58.7%, compared with 30.1% for its base model and 55.4% for the 2‑billion‑parameter competitor. Across the six evaluations, BotPsy’s average score was 45.0%, versus 38.8% for the larger Qwen model.
Agentic capabilities
Unlike typical chatbots, BotPsy is described as “agentic,” meaning it can identify required tools, execute commands in sequence and adapt when errors occur. In a benchmark scenario, the assistant detected that Wi‑Fi was disabled and the inbox was full, turned on Wi‑Fi, cleared storage space and successfully sent a message, whereas a comparable model repeatedly failed.
Integration with Tether’s QVAC ecosystem
BotPsy expands Tether’s QVAC ecosystem, which includes locally run AI models for medical applications, translation and speech processing. The model is released under the Apache 2.0 license, with compressed versions available for lower‑memory devices. Tether has not announced a consumer‑ready assistant built on BotPsy, and the release does not include the training code or data.
