SCBX
Weekly Tech
New Framework Doubles Success Rate for Long-Horizon AI Agents
Researchers introduced StructAgent, a framework that replaces raw interaction-history processing with a unified, verifiable state representation for long-horizon digital agents. The approach enables progress checkpointing, evidence-driven task completion, and targeted failure recovery. On the OSWorld-Verified benchmark, it roughly doubled task success for Qwen3.5-9B (27.0% to 46.9%) and Qwen3.5-27B (31.6% to 62.2%), and the authors showed the method generalizes beyond desktop automation to Minecraft environments.
Connect with us
via LINE OA.
Stay updated with R&D. Scan the QR.

