Tether launches 800000000-parameter on-device AI agent model QVAC BotPsy
BIBIBI
AT A GLANCE
Tether launches the 800000000-parameter open-weight model QVAC BotPsy, which can run locally on personal devices and convert user requests into tool calls.
Article
Tether Releases 800000000-Parameter Open-Source On-Device AI Agent Model QVAC BotPsy
PANews November 11 News, according to Tether's official announcement, its AI research department has launched the open-weight model QVAC BotPsy, with a parameter count of 800000000. Developed based on Alibaba's open-source Qwen3.5 architecture, it can run locally on personal devices such as smartphones, converting user requests into tool calls to help users handle daily tasks like scheduling and messaging. Tether stated that among the 6 agent benchmarks tested, BotPsy outperformed all models under test with no more than 1200000000 parameters; in four core tests—BFCL v4, ACEBench, τ-bench, and ToolSandbox—it surpassed Qwen3.5-2B-Think, which has approximately 2.5 times its parameter count, achieving tool-calling scores roughly twice those of its base model. The model weights are now available on Hugging Face.
According to PANews's relay of Tether's official announcement, Tether's AI research division has launched QVAC BotPsy, with a parameter count of 800000000.
02
The model is developed based on Alibaba's open-source Qwen3.5 architecture and can run locally on personal devices such as mobile phones, handling everyday tasks like schedules and messages.
03
Tether stated that BotPsy outperformed all tested models with no more than 6 parameters in its tests across 1200000000 agent benchmarks.
04
Tether stated that in four core tests—BFCL v4, ACEBench, τ-bench, and ToolSandbox—BotPsy outperformed Qwen3.5-2B-Think, with a tool-calling score approximately twice that of its base model.
05
The report stated that the model weights have been made available on Hugging Face.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This is an AI model that converts user requests into tool operations, for example assisting with schedules and messages. The materials state that it can run directly on personal devices such as phones; the performance advantage comes from test results published by Tether and cannot be directly equated to actual usage results.
Why it matters to readers
This release showcases the direction of running AI task assistants on personal devices, with a focus on whether smaller parameter models can effectively perform tool calling.
It can be understood as a local AI assistant model capable of assisting with tool operations, but open model weights do not equate to having already provided a directly usable mobile application.
Risks and unknowns
Performance comparisons are as described by Tether; the materials do not provide independent verification results or complete test conditions.
The material does not specify the specific devices supported, the runtime resource requirements, or the actual task performance.
The materials do not provide a specific license, so it is not possible to confirm the full open-source scope and usage restrictions based solely on open weights.
Related Developments
Loading event timeline…
Related concepts
Parameter count
This term is not in the glossary yet. Browse related concepts in the glossary.
