Tether AI Research released QVAC BotPsy on November 11, 2026, adding a 0.8-billion-parameter open-weight model to the company’s expanding artificial-intelligence portfolio. Built on Alibaba’s open-source Qwen3.5 architecture and fine-tuned for agentic tasks, BotPsy converts user requests into tool calls for everyday activities such as managing schedules and messages. It outperformed all tested models with no more than 1.2 billion parameters across six agentic evaluations. In four core tests—BFCL v4, ACEBench, τ-bench and ToolSandbox—it also surpassed Qwen3.5-2B-Think, which has roughly 2.5 times as many parameters, while achieving tool-calling scores approximately twice those of its base model. Earlier testing recorded a mean score of 45.0 across six evaluations, including 44.0 on BFCL v4 versus 23.4 for its base model and 51.5 on τ-bench versus 23.2. Its BFCL v4 hallucination-resistance score was 86.1, the highest among competitors below 1.2 billion parameters. The Apache 2.0-licensed model is available on Hugging Face, including GGUF formats for local hardware, and is intended primarily for research, education and low-to-medium difficulty tasks. Its ability to operate on personal devices such as smartphones supports Tether’s QVAC initiative to develop decentralized AI without cloud dependence. The results reinforce the push toward more efficient small models, although the narrow six-test scorecard and Tether’s usage guidance position BotPsy as a focused tool-use model rather than a general replacement for larger systems. The release introduces no token or new tradable asset.