Opus 5 code review evals: cleaner actionable comments but noisier overall (www.coderabbit.ai)

Anthropic has launched Claude Opus 5, a significant update in its Opus series, which aims to enhance code reviews in the AI/ML community. While the update doesn't dramatically increase bug detection rates, it shifts focus towards generating cleaner, more actionable comments. Opus 5 x-high configuration stands out by producing 39.3% actionable comments compared to the previous 35.2%, but it also caught fewer known issues (55.2% vs. 61.1%) and delivered roughly four times as many minor feedback points (nitpicks). The model is recommended as a supplementary reviewer where precision is paramount, particularly in projects requiring detailed attention. Despite its insights into integration errors and code quality, Opus 5 falls short in identifying critical issues like logic errors and API misuse. The model's design improves collaboration and exploration of solutions in complex coding tasks, making it more effective for ambiguous projects than its predecessor, Opus 4.8. However, compared to competitors like Fable 5, Opus 5 is found to be less efficient, taking longer to deliver outputs while managing a larger volume of input tokens. The key takeaway is that while Opus 5 excels in generating clear comments, its increased noise levels necessitate careful filtering before being used in critical code reviews.