Tether unveils 0.8B AI model for on-device assistants

Traders Union
All News
Crypto
Tether unveils 0.8B AI model

Highlights

Tether AI Research launched QVAC BotPsy, a sub-1B parameter on-device model doubling tool-use scores over Qwen3.5-0.8B-Think, with BFCL v4 rising from 23.4 to 44.0.
BotPsy outperforms all tested models at 1.2B parameters or below across six benchmarks, including beating Alibaba’s larger Qwen3.5-2B-Think in four primary agentic evaluations.
The on-device BotPsy model targets tasks like messaging and scheduling, aiming to increase user privacy by processing sensitive data locally instead of relying on external cloud providers.

Benchmark gains and product positioning

As reported by Tether News, Tether AI Research announced QVAC BotPsy, a sub-1B parameter model built to convert user requests into tool calls across apps and services. The company positions the release as part of its QVAC strategy for personal assistants that can manage everyday tasks without relying on centralized cloud processing.

BotPsy is built on Alibaba’s open-source Qwen3.5 architecture and is designed for on-device deployment. Tether says the model roughly doubles the tool-use scores of its Qwen3.5-0.8B-Think base architecture on key evaluations, with BFCL v4 improving from 23.4 to 44.0, ACEBench from 30.1 to 58.7, τ-bench from 23.2 to 51.5, and ToolSandbox from 30.5 to 64.6.

Across six benchmarks, BFCL v4, ACEBench, τ-bench, Claw-Eval, BrowseComp-Plus and ToolSandbox, the company says BotPsy leads comparable models and beats all tested models at 1.2B parameters or below. On four primary agentic evaluations, Tether says it also outperforms Alibaba’s Qwen3.5-2B-Think, a model 2.5 times larger.

The strongest development gap appears in BFCL’s multi-turn tool-calling test, which measures how a model continues tasks across successive user interactions. BotPsy scores 46.1 versus 23.8 for Qwen3.5-2B-Think, while the underlying base model scores 7.5.

On-device AI strategy and industry implications

BotPsy is aimed at assistant use cases such as sending messages, rescheduling meetings and coordinating reminders, where systems need access to emails, calendars, contacts and travel plans. Tether argues that a small model running locally can process those requests on personal devices rather than sending sensitive information to servers controlled by large technology providers.

In one ACEBench example cited by the company, BotPsy turns Wi-Fi back on and frees inbox space before sending a failed message, while its base model keeps retrying the same action. Tether also says the model can identify when no available tool fits a request, instead of inventing a tool call, a behavior that is important for reliable consumer assistants.

Paolo Ardoino, CEO of Tether, says local AI can give users more control over how personal information is processed and which external services their assistants can access. The company says BotPsy joins VisionPsy, TranscribePsy, MedPsy and TranslatePsy in its QVAC ecosystem, with a longer-term plan to use smartphones as the daily interface while home PCs or trusted peer-to-peer nodes handle more demanding tasks through encrypted connections.

We have also looked at Lenovo’s Hybrid AI strategy, which highlights privacy, security and data control across consumer and enterprise systems.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.