Key facts
- Tether unveiled QVAC BotPsy, an 800-million-parameter AI agent for tasks on personal devices.
- In Tether's testing, BotPsy beat models with up to 2.5 times as many parameters on five of six tool-use benchmarks.
- BotPsy scored 44.0% on BFCL v4, against 23.4% for its base model and 37.7% for Qwen3.5-2B.
- BotPsy is built on Alibaba's Qwen3.5-0.8B architecture and further trained by Tether for software tool use.
- BotPsy is available as an open-weight model under the Apache 2.0 license, with compressed versions for local deployment.
- Tether has not announced a release date for a finished consumer assistant powered by BotPsy.
Tether unveiled QVAC BotPsy on Wednesday, an 800-million-parameter AI agent built to carry out everyday tasks on personal devices, according to a report published November 11, 2026. Parameters are the numerical values a model learns during training; more of them generally mean more capability, and frontier models run in the trillions.
In Tether's own testing, BotPsy outperformed models with up to 2.5 times as many parameters. The company tested five models across six tool-use evaluations, and BotPsy posted the highest result on five, with an average score of 45.0% against 38.8% for the larger Qwen model.
On BFCL v4, which measures an AI system's ability to select and execute software functions across tasks and conversations, BotPsy scored 44.0%, per Tether's technical documentation. The underlying Qwen3.5-0.8B model scored 23.4% and the larger Qwen3.5-2B reached 37.7%. On ACEBench, which evaluates the accuracy of function calls including tool and argument selection, BotPsy scored 58.7%, its base model 30.1% and the 2-billion-parameter competitor 55.4%.
BotPsy is built on Alibaba's Qwen3.5-0.8B architecture, which Tether trained further to specialize in interacting with software tools. Unlike chatbots that respond to questions, an agentic model is designed to take action: identify which tools an application needs, execute commands in order and adjust when something fails. Tether described a benchmark scenario in which an assistant needed to send a message but found Wi-Fi disabled and the inbox full; BotPsy turned on Wi-Fi, cleared storage and completed the task while the underlying model repeated the same failed operation.
BotPsy joins QVAC, Tether's effort to run AI on devices controlled by users rather than relying entirely on centralized cloud infrastructure. Tether introduced QVAC in May 2025 and expanded it in October 2025 with a 41-billion-token dataset for training science and engineering models and QVAC Workbench, an application for running models locally. It also released an SDK for offline AI applications and developed QVAC MedPsy, a medical system whose 1.7-billion-parameter version outperformed Google's MedGemma-27B on HealthBench Hard, a test of clinical conversations evaluated using physician-developed criteria, according to the company.
BotPsy is available as an open-weight model under the Apache 2.0 license, letting developers download, modify and integrate it, and Tether has released compressed versions to reduce memory needs for local deployment. Tether has not announced a release date for a finished consumer assistant powered by BotPsy, and its model card says training code and training data are not included.