Tether's QVAC BotPsy, a 0.8B model, beats larger rivals on tool-use tests, company says
In brief
- Tether AI Research released QVAC BotPsy, a 0.8-billion-parameter open-weight model, on November 11, 2026.
- BotPsy scored a mean of 45.0 across six agentic evaluations, Tether reports.
- BotPsy beat Qwen3.5-2B-Think on five of six tests, according to Tether's results.
- BotPsy ships under Apache 2.0 in GGUF formats for running on local hardware.
- No token or new tradable asset is attached to the release.
What Tether's numbers show
Tether says BotPsy now sets the state of the art in the sub-1B parameter class on key tests of how well AI agents use APIs and tools. These are the company's own reported results (relayed by Crypto Briefing), and LeoDex News hasn't independently verified them.
The headline figure is a mean score of 45.0 across six agentic evaluations. Tether says BotPsy beat Qwen3.5-2B-Think, a 2-billion-parameter model, on five of those six tests. It didn't win all six.
The gains over the base model are bigger. On the BFCL v4 function-calling benchmark, Tether reported a score of 44.0, up from 23.4 for the base model; on τ-bench, BotPsy reached 51.5 compared with 23.2. Tether also reported a hallucination resistance score of 86.1 on BFCL v4, which it calls the top result among competitors under 1.2B parameters.
Crypto Briefing noted that six evaluations is "a focused scorecard, not a full report card."
Built to run on hardware users control
BotPsy was fine-tuned from Qwen3.5-0.8B using a range of learning strategies to improve its agentic skills. Tether released it under the Apache 2.0 license and shipped it in GGUF formats, a file type designed for running models on local hardware.
The local angle is the whole point. BotPsy is part of Tether's broader QVAC initiative, which aims at decentralized AI that runs entirely on local devices (no cloud dependence). Tether describes the release as advancing its push to put personal assistants for everyday tasks onto hardware users control.
BotPsy joins the QVAC Psy family, which already includes VisionPsy and MedPsy.
The caveats
Tether isn't pitching this as a do-everything model. The company positions BotPsy primarily for research and educational purposes, and it says the model is best suited to low-to-medium difficulty tasks.
There's also no token here. The launch carries no token mechanics and introduces no new tradable asset.
Frequently asked questions
What is Tether's QVAC BotPsy?
QVAC BotPsy is an open-weight AI model with 0.8 billion parameters, released by Tether AI Research on November 11, 2026. It was built on Alibaba's Qwen3.5-0.8B base model and fine-tuned for agentic tasks such as email function calling and document handling.
Can BotPsy run on local hardware?
Yes. Tether shipped BotPsy in GGUF formats, a file type designed for running models on local hardware, under the Apache 2.0 license. It's part of Tether's QVAC initiative, which aims at decentralized AI that runs on local devices without relying on the cloud.
Did BotPsy beat larger models on every benchmark?
No. According to Tether's reported results, BotPsy beat Qwen3.5-2B-Think on five of six agentic tests, not all six. Tether says the model is best suited to low-to-medium difficulty tasks and positions it mainly for research and educational use.
