ARCHIVE EDITION2026-11-16
TODAY'S STYLEIslamic Geometric

BotPsy puts tool calls in a 0.8B on-device model

Tether AI Research released BotPsy-0.8B on November 11, adapting Qwen3.5-0.8B for function calls and agent workflows. Its model card reports leading results against the tested models with up to 1.2 billion parameters across six agent benchmarks. Those are developer-reported comparisons, rather than an independent ranking of every model. Apache 2.0 weights support local deployment, with English recommended. The intended workload is low to medium difficulty. A small local model still makes mistakes, so tool permissions, approval steps and device latency require separate testing before practical use. Local execution alone does not establish safe operation.

SourceTether AI Research