Claude Opus 4.8 posts 69.2% on SWE-Bench Pro, extending its lead in agentic coding benchmarks, yet still trails GPT-5.5 on Terminal-Bench 2.1 at 2.1. The model now admits uncertainty on select tasks, a shift from prior versions that rarely flagged their own limits. Released at unchanged pricing, the update arrives alongside EasyRouterIO launch offering 400 free test credits via promo code.