Opus 4.7 Long-Context Regression

In a system card comparison, Opus 4.7 improved on the SWE-bench series, but Opus 4.6 was superior on the remaining 5 items.

• SWE-bench Verified: 80.8% → 87.6%
• SWE-bench Pro: 53.4% → 64.3%

On the other hand, the search and long-context reasoning series declined.

• BrowseComp (10M token): 83.7% → 79.3%
• DeepSearchQA F1: 91.3% → 89.1%
• MRCR v2 8-needle @ 256k: 91.9% → 59.2%
• MRCR v2 8-needle @ 1M: 78.3% → 32.2%
• ARC-AGI-1: 93.0% → 92.0%

The drop is especially large in MRCR v2 @ 1M, which reads as a signal that 4.7's ability to accurately find specific information in long-context has weakened significantly.