Tether’s 800M AI Agent Outperforms Models 2.5 Times Its Size in Six Tool-Use Tests

11/11/2026 admin

Key Takeaways:

Tether’s QVAC BotPsy-0.8B beat models with up to 2.5 times more parameters in six agentic tests.
The tool-use score of the 800M-parameter model almost doubled the scores of its base Qwen3.5-0.8B-Think architecture.
To move more powerful AI assistants to personal devices and put them away from centralized servers, Tether plans to do so.

User Score

8.7

Tether has introduced QVAC BotPsy, an open-weight AI model designed to execute tasks through apps and external tools. The model, with only 0.8 billion parameters, beat bigger models on tests arranged by Tether, suggesting the ability of tailored AI to serve daily needs on individual gadgets.

Tether’s 800M AI Agent Beats Models 2.5x Its Size on Everyday Phone Tasks
Learn more: https://t.co/3GFAZrkM9W
— Tether (@tether) November 11, 2026

Tether’s 800M Model Challenges Larger AI Systems

After the announcement on November 11, QVAC BotPsy outperformed all models with 1.2 billion or fewer parameters on six respective agentic benchmarks: BFCL v4, ACEBench, τ-bench, Claw-Eval, BrowseComp-Plus, and ToolSandbox.

On four key metrics, BotPsy almost doubled the scores given by its base model Qwen3.5-0.8B-Think. It improved its score for BWFC v4 from 23.4 to 44.0 and ACEBench from 30.1 to 58.7. It also raised its scores on the τ-bench and ToolSandbox benchmarks up to 51.5 and 64.6, respectively.

Tether also claims that BotPsy does outperform Alibaba’s model Qwen3.5-2B-Think, which has 2.5 times the number of parameters. The base model achieved 46.1 points and the larger model 23.8 points on BFCL multi-turn tool calling, whereas its base architecture got 7.5 points.

BotPsy significantly outpaced MiniCPM5-1B-Think with a score of 82.5, outshone Qwen3.5-2B-Think with 69.3 and edged out LLama3-3B-Think at 74.4.

BotPsy Turns Everyday Requests into Actions

BotPy is not built for generation of responses only like AI models are, but it is built to call the right tool calls in multiple steps. This might help with organizing activities, sending messages, and even travel arrangements.

One ACEBench test attempted to send a message, but it was unable to do so due to a lack of Wi-Fi connectivity and a full inbox. BotPsy spotted both, activated WiFi, removed inbox space and then sent the message. Instead, its bottom line will try the same action repeatedly that it has failed on.

The model can define when there may be no available tool needed to address a request as opposed to making an unwanted Call. This is significant for assistants that must perform numerous actions and modify content.

Local AI Could Reduce Reliance on Centralized Servers

BotPsy is based on Alibaba’s open-source Qwen3.5 and is sufficiently small to allow it to be explored on personal hardware. Requests could be satisfied locally, limiting the amount of sensitive information: emails, contacts, calendars, and more, sent to central providers.

CEO Paolo Ardoino noted that with local AI, users will have more control over the processing of their personal data and the accessibility of services for their assistant.

BotPsy extends the QVAC ecosystem developed by Tether including VisionPsy which tackles Visual Question Answering, TranscribePsy which addresses speech and TranslatePsy which addresses translation tasks, etc.

The post Tether’s 800M AI Agent Outperforms Models 2.5 Times Its Size in Six Tool-Use Tests appeared first on CryptoNinjas.

CryptoNinjas