AI · Studies · 2026-09-09 11:51:00
Sierra open-sourced hyper-𝜏-bench (published as 𝜏^𝜏-bench), a new long-horizon agent evaluation measuring how well models can not only act as an agent but construct one. A developer agent recovers requirements from a business's records, builds a customer-service agent in a sandboxed workspace, and is scored on how that agent handles held-out tasks. Working alone, the best configuration—Claude Opus 5 in Claude Code—passes just 23.9% of held-out tasks; paired with an engineer with deep context, the same class of model reaches 82.2%.
Auto-generated summary for reference only. RSS