The stablecoin issuer's 0.8-billion-parameter model beat rivals up to 2.5 times its size on tool-use tests built for AI agents
The company best known for issuing the USDT stablecoin now has a new item on its résumé: a tiny AI model that beats bigger ones at their own game.
Tether AI Research released QVAC BotPsy on November 11, 2026. It is an open-weight model with 0.8 billion parameters, and on tool-use benchmarks it outperformed models up to 2.5 times its size.
Small model, big scorecard #
BotPsy runs with 0.8 billion parameters. That puts it in the sub-1B class, where Tether says the model now sets the state of the art across key tests of how well AI agents use APIs and tools.
An “agent” here means an AI that does things rather than just chats. It calls functions, fills in forms, sorts email and manages documents. The model has to pick the right tool, feed it the right inputs and avoid wandering off into fiction.
Across six agentic evaluations, BotPsy posted a mean score of 45.0. It beat Qwen3.5-2B-Think, a 2-billion-parameter model, on five of those six tests.
The gains over its own starting point were steep. On BFCL v4, a function-calling benchmark, BotPsy scored 44.0, up from 23.4 for its base model. On τ-bench, it reached 51.5, compared with 23.2 for the base.
BotPsy recorded a hallucination resistance score of 86.1 on BFCL v4, the top result among competitors under 1.2B parameters.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Built on Qwen, tuned by Tether #
BotPsy did not start from scratch. It is built on Alibaba’s Qwen3.5-0.8B base model and was then fine-tuned using a range of learning strategies to sharpen its agentic skills. Tether’s job of choice is everyday digital chores, including email function calling and document handling.
The release comes under the Apache 2.0 license, a permissive open-source license that allows broad use and modification. Tether also shipped the model in GGUF formats, a file type designed for running models on local hardware.
Tether positions BotPsy primarily for research and educational purposes. The company says the model is best suited to low-to-medium difficulty tasks.
BotPsy joins the QVAC Psy family, which already includes models such as VisionPsy and MedPsy.
The QVAC bet on local AI #
BotPsy is part of Tether’s broader QVAC initiative. The goal is decentralized AI that runs entirely on local devices, with no dependence on the cloud.
Tether describes the release as advancing its push to put personal assistants capable of handling everyday tasks onto hardware that users control.
What this means #
For the AI industry, BotPsy is another data point in a quiet but meaningful trend: efficiency gains in small models. If a 0.8-billion-parameter model can beat a 2-billion-parameter rival on most agentic tests, the hardware bar for useful on-device assistants drops. The benchmark results deserve some perspective, though. Six evaluations is a focused scorecard, not a full report card, and BotPsy did not win all six head-to-heads against Qwen3.5-2B-Think. Tether’s own guidance limits it to lower-difficulty work. Builders should treat it as a strong option for narrow, well-defined agent tasks rather than a general replacement for larger systems.
For Tether, the release adds substance to a strategy that extends well beyond stablecoins. The launch carries no token mechanics, so there is no new asset to trade here. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.