Tether’s 800M-Parameter BotPsy Targets Local AI Agent Tasks

Tether AI Research has released BotPsy-0.8B, an open-weight AI model built for local agents that can carry out multi-step tasks on phones and interact with tools. The model contains 800 million parameters and is intended for everyday device-based workflows rather than requiring every operation to run through a cloud service, according to Crypto Economy and Decrypt. Crypto Economy reports that BotPsy outperformed Qwen3.5-2B-Think in Tether’s testing even though the rival model is 2.5 times larger. It also achieved roughly twice the score of its underlying base model across four core agentic benchmarks. Crypto Economy further says BotPsy led the tested systems on a BFCL hallucination assessment, although the supplied coverage does not specify the comparative figures for that result. The two outlets corroborate the model’s 800-million-parameter scale, its open-weight release and its focus on local agentic work. Crypto Economy concentrates on benchmark performance and efficiency relative to larger models, whereas Decrypt presents BotPsy as a practical everyday assistant and connects it to Tether’s expansion into decentralized artificial intelligence. No material contradiction appears between their coverage. The release extends Tether’s activity beyond its stablecoin business into compact models designed for autonomous, tool-using applications. BotPsy’s emphasis on local execution and multi-step phone tasks positions it for users seeking agent capabilities without placing the entire workflow on a remote server, a design choice highlighted by both Crypto Economy and Decrypt. Live market context shows USDT at $0.999166, unchanged over 24 hours and down 0.01% over seven days, with a market capitalization of $175.81B. DeFiLlama verifies a slightly different circulating-supply figure of $175.67B. The broader crypto market capitalization stands at $2.54T, Bitcoin is up 0.91% over 24 hours, and the Fear & Greed index reads 29, categorized as Fear. The market backdrop remains relatively flat despite the AI-model announcement.