Jan 30, 2026 14:00:00

'Claude Code Opus 4.5' is degraded

AI provided by companies such as OpenAI and Anthropic is constantly being updated, and performance and output trends can change even for AI models with the same name.

Marginlab , a company that measures AI performance, conducted a follow-up study and found that the performance of Claude Code Opus 4.5 had deteriorated.

Claude Code Opus 4.5 Performance Tracker | Marginlab
https://marginlab.ai/trackers/claude-code/

Marginlab uses the benchmark test ' SWE-Bench Pro ' to measure the performance of Claude Code Opus 4.5 daily and analyze changes in performance. The graph below shows the progress of the score up to January 29, 2026. Marginlab has issued an alert indicating that 'degradation has been observed' based on the progress of the score over the past 30 days.

The Claude Code Opus 4.5 score was down 8.0% from the previous day, down 4.8% from the previous week, and down 4.1% from the previous month, with the month-on-month data being statistically significant.

Marginlab has also

performed a similar follow-up study on OpenAI's Codex gpt-5.2-high, but no significant changes have been observed.

There are no statistically significant differences compared to the previous day, week, or month.

The information that 'Claude Code Opus 4.5 performance is deteriorating' has become a hot topic on the news sharing site Hacker News, and Thariq Shihipar, who is in charge of Claude Code development, appeared in the thread and explained , 'We encountered a problem with the Claude Code harness on January 26, 2025. We discovered the problem on January 28, 2026 and performed a rollback.'