Anthropic's Opus 5.5 Drops 9.6 Points on NerfBench, Fueling Persistent Model-Degradation Concerns

Claude Opus 5.5 fell sharply to 94.2% of its launch performance on NerfBench as of October 2, marking the largest single-day decline since Anthropic's flagship reasoning model debuted nine days earlier. The drop—from 103.8% on October 1—reignited developer concerns about model degradation after release, even as competing models from OpenAI and Anthropic itself demonstrate stable or improving performance.