Was Opus 5.5 nerfed? Looks like it wasn't, according to LiveNerf (blog.nathanlangley.dev)

A new project called LiveNerf has emerged to investigate claims of AI models, particularly Claude, becoming less effective over time—often referred to in the community as "nerfing." Developed by an AI enthusiast, LiveNerf continuously benchmarks models to provide ongoing performance evaluations, rather than just a snapshot at launch. The initial findings reveal that Opus 5.5 shows no significant decline in performance, with scores fluctuating but maintaining an average close to its initial benchmark, thus countering the narrative that the model has been negatively impacted. This continuous evaluation is critical for the AI/ML community, as it provides transparency regarding model performance post-release—an area often overlooked. While many complaints about models relate to varying interaction contexts or system adjustments rather than inherent model degradation, LiveNerf's findings underscore the need for a robust method to discern between actual regressions and normal usage variations. As the project evolves, it aims to refine its benchmarks and expand its monitoring capabilities, highlighting the importance of consistent evaluation in a field where AI is increasingly relied upon for complex tasks.