Tether releases QVAC BotPsy, an 800-million-parameter AI model
- Tether
- QVAC
- BotPsy
- on-device AI

The model scored 44.0% in initial
BFCL v4resultsQVAC tools are available for running models on consumer devices
Tether is expanding QVAC from a local-first AI concept into an open-source ecosystem with tools developers can use to run and adapt models on consumer devices. The company says its approach is designed to keep AI processing on laptops, desktops and phones rather than relying solely on remote data centers.
Tether first announced QVAC in May 2025 and launched the QVAC SDK on April 9, 2026. A software development kit, or SDK, provides tools for building applications; QVAC’s cross-platform inference runtime is intended to run the same model on different types of devices without rebuilding it for each one. Inference is the process in which a model produces an answer.
QVAC Fabric supports fine-tuning models on consumer hardware, including mobile graphics processing units. Fine-tuning adapts a general model for a more specific purpose. Fabric uses LoRA, a technique that updates a smaller part of a model instead of retraining it entirely. Tether has also released core components under permissive open-source licenses, allowing outside developers to inspect and build on the code.
The latest model in the ecosystem is QVAC BotPsy, which Tether released on November 11, 2026. The company describes it as an 800-million-parameter model optimized for agentic tasks, in which an AI uses tools or performs multiple steps to complete a task rather than only holding a conversation.
In initial benchmarks cited in the reporting by Crypto Briefing, BotPsy scored 44.0% on BFCL v4, compared with 37.7% for a model with 2 billion parameters. Across a broader set of tests, the article reports that BotPsy outperformed models up to 2.5 times larger in five of six benchmarks. These are benchmark results, not evidence that BotPsy performs better across every task or setting. Tether’s QVAC lineup also includes VisionPsy, for which the article gives no benchmark scores.
QVAC is linked to Tether’s Wallet Development Kit, which supports transactions in Bitcoin and USDT. Tether’s broader plan is to incorporate payment capabilities into AI agents. The reporting does not establish that QVAC agents already make payments as a general-purpose, deployed feature.
Running a model locally can mean that prompts do not need to be sent to a centralized cloud service, a potential privacy benefit for users and developers. It does not by itself establish that an application sends no other data. Tether has described an ambition to support trillions of agents on everyday devices, combining local inference, decentralized networking and payment integration; the article presents that as a goal, not a current deployment at that scale.
Key terms explained
- QVAC
QVAC is an open-source local AI ecosystem designed to run models on devices such as phones and laptops.
- LoRA
LoRA is a method that adapts a model by training low-rank matrices while keeping its pre-trained weights fixed.
- inference
Inference is the process of using a model to produce output from input.
Our priority is to report the news promptly and faithfully.
We use AI to summarize the key facts in our sources. Our editors are responsible for final review and publication.
We leave it to readers to interpret the facts and draw their own conclusions.













