AI·yesterdayClaude Opus 4.7 is here, and the long-context benchmarks got worseAnthropic's Opus 4.7 is state-of-the-art on SWE-bench and CursorBench, but independent tests show regressions on long-context retrieval and thematic reasoning.
