Tether, the company behind USDT, the world's largest stablecoin, has unveiled a compact AI agent that outperformed models up to 2.5 times its size in the company's own benchmark testing, a sign that capable AI assistants may not need enormous models to get things done.

On Wednesday, the company announced QVAC BotPsy, an 800-million-parameter AI agent designed to carry out everyday tasks on personal devices, including managing phone settings, messaging and scheduling, potentially without sending personal information to cloud servers.

Parameters are essentially the knobs that determine how much an AI model can do, and more of them generally translates into improved performance. For reference, frontier AI models today run in the trillions of parameters. Yet in company testing, Tether's relatively tiny model beat competitors with up to 2.5 times as many parameters on five of six tool-use benchmarks.

Size is more than a scoreboard quirk for this kind of software: consumer hardware imposes strict memory limits, so the parameter count of a model meant to run locally shapes what can realistically operate on the device in someone's pocket.

An agent, not a chatbot

Unlike traditional chatbots that simply respond to questions, BotPsy is agentic, meaning it is designed to take action. It can identify which tools an application needs, execute commands in the right order, and adjust when something goes wrong.

Tether described a benchmark scenario in which an AI assistant needed to send a message but encountered two problems: Wi-Fi was disabled and the inbox was full. BotPsy switched on Wi-Fi, cleared enough storage and completed the task, while the underlying model repeatedly attempted the same failed operation. It is a fairly mundane achievement, but precisely the kind of problem that makes AI agents useful outside a chatbot window.

Built on a small Qwen model

BotPsy is built on Alibaba's Qwen3.5-0.8B architecture, a relatively small language model that Tether trained further to specialize in interacting with software tools. Its 800 million parameters represent the numerical values the model learned during training, and a smaller parameter count generally means less memory is needed to run it.

According to Tether's technical documentation, BotPsy scored 44.0% on BFCL v4, a benchmark that evaluates an AI system's ability to select and execute software functions across different tasks and conversations. Scores reflect a weighted percentage of successful performance across tool-use categories. The underlying Qwen3.5-0.8B model scored 23.4%, while the considerably larger Qwen3.5-2B reached 37.7%.

BotPsy also scored 58.7% on ACEBench, which evaluates the accuracy of function calls, including selecting the appropriate tools and arguments. Its base model managed 30.1%, while the 2-billion-parameter competitor scored 55.4%.

The company tested five models across six tool-use evaluations, with BotPsy posting the highest results on five. Its average score across those tests reached 45.0%, compared with 38.8% for the larger Qwen model.

Part of the QVAC ecosystem

BotPsy is the latest addition to QVAC, Tether's effort to build an AI ecosystem that runs on devices controlled by users rather than relying entirely on centralized cloud infrastructure. The company first introduced QVAC in May 2025, positioning the project as an alternative to AI services operated by technology giants. Its longer-term ambitions included autonomous agents capable of interacting with applications and potentially making payments using Bitcoin and USDT.

That payments ambition is also where Tether's AI program meets its core business: an agent able to transact in USDT would be handling the very stablecoin the company issues.

Tether expanded the initiative in October 2025 with a 41-billion-token dataset for training science and engineering models, alongside QVAC Workbench, an application that lets users run AI models locally. Earlier this year, the company released its software development kit, providing developers with tools to build offline AI applications across mobile and desktop operating systems.

For that vision, compact models are a practical prerequisite, since the ecosystem's premise is that AI should run on hardware users control rather than in distant data centers.

Tether has also developed QVAC MedPsy, a medical AI system designed to run on consumer hardware. Its 1.7-billion-parameter version outperformed Google's much larger MedGemma-27B on HealthBench Hard, a test of realistic clinical conversations evaluated using physician-developed criteria.

Availability and caveats

BotPsy is currently available as an open-weight model under the Apache 2.0 license, meaning developers can download, modify and integrate it into their own applications. Tether has also released compressed versions intended to reduce memory requirements for local deployment.

The company has not announced a release date for a finished consumer assistant powered by BotPsy, and its published model card notes that the training code and training data are not included in the release.

That leaves developers as the immediate audience for the release: the open weights and compressed builds are already downloadable, while any finished consumer assistant built on BotPsy remains undated.