[Tether Releases 800 Million Parameter End-Side Model BotPsy, Claims Tool Call Performance Surpasses Models 2.5 Times Larger] BlockBeats reported on November 11th that Tether announced the release of the open-weighted model QVAC BotPsy, with 800 million parameters. Based on Alibaba's open-source Qwen3.5 architecture, it can transform everyday requests into local tool calls on users' own devices. Tether claims that in its own testing, the model achieved a new high in tool call evaluations for models with fewer than 1 billion parameters, and surpassed models with 1.2 billion parameters and below. In four major benchmarks—BFCL v4, ACEBench, τ-bench, and ToolSandbox—its performance was roughly twice that of the base model Qwen3.5-0.8B-Think, and also surpassed Qwen3.5-2B-Think, which has 2.5 times the number of parameters. Specifically, BFCL achieved a score of 46.1 in multi-round tool calls, approximately twice the 23.8 score of the 2B model. Tether explains that BotPsy can reduce the amount of personal information such as emails and calendars sent to centralized service providers, and can also determine when no suitable tools are available, preventing fabricated calls. Tether CEO Paolo Ardoino stated that users should decide for themselves what content the assistant can access and what external services it can connect to. The model weights have been published on Hugging Face. [Original Link]