Tether launches 800M‑parameter BotPsy AI agent that outperforms larger rivals
Tether’s new BotPsy model, built on an 800‑million‑parameter architecture, beat larger competitors on five of six tool‑use benchmarks, showcasing the potential of on‑device AI.
BotPsy uses 800 million parameters and outperformed models up to 2 billion parameters on five of six tool‑use tests.
The model scored 44.0% on BFCL v4 and 58.7% on ACEBench, surpassing the base Qwen3.5‑0.8B and Qwen3.5‑2B scores.
BotPsy is released under an Apache 2.0 license with compressed versions for local deployment on mobile devices.
Tether announced the release of BotPsy, an 800‑million‑parameter AI agent designed to execute multi‑step tasks on personal devices. In internal testing, the model outperformed larger rivals on five of six tool‑use benchmarks, suggesting that compact, on‑device AI can rival much bigger systems.
BTC / USD
Source: CoinGecko. Prices are indicative.
Benchmark performance
According to Decrypt, BotPsy achieved a 44.0% score on the BFCL v4 benchmark, compared with 23.4% for its underlying Qwen3.5‑0.8B model and 37.7% for the 2‑billion‑parameter Qwen3.5‑2B version. On the ACEBench test, which measures the accuracy of function calls, BotPsy posted 58.7%, beating the base model’s 30.1% and the larger model’s 55.4%. Across the six evaluations, BotPsy’s average was 45.0%, outpacing the 38.8% average of the larger Qwen model.
Agentic capabilities and on‑device focus
Unlike typical chatbots, BotPsy is “agentic,” meaning it can identify required tools, execute commands in sequence, and recover from errors. Decrypt cited a scenario where the assistant turned on Wi‑Fi, cleared storage, and sent a message after encountering two obstacles. The model runs locally, reducing reliance on cloud servers and limiting the transmission of personal data.
Integration with Tether’s QVAC ecosystem
BotPsy extends Tether’s QVAC platform, which already includes offline models for medical analysis, translation and speech processing. The company first unveiled QVAC in May 2025 and has since added a 41‑billion‑token dataset for scientific models and an SDK for developers. BotPsy is offered as an open‑weight model under the Apache 2.0 license, with compressed builds aimed at low‑memory mobile hardware.
What to watch next
Tether has not set a launch date for a consumer‑facing assistant built on BotPsy, but the open model may spur third‑party integrations. Observers will monitor whether developers adopt the SDK to create privacy‑preserving AI tools and whether Tether expands BotPsy’s functionality to include payment actions using USDT or Bitcoin.
Kenji SatoKenji Sato is a blockchain analyst. He explains how new networks and protocols work: Ethereum and layer 2s, DeFi, upgrades, mining and the technology behind the projects in the news.All articles by the author
Discussion 0
Be the first to weigh inWhat does this mean for the market? Share your take and get the discussion going.
Discussion 0