ThinkingNews

Back to feed
TechMeme July 8, 2026

OpenAI says it found widespread task issues in SWE-Bench Pro, estimates ~30% of tasks are broken, and retracts its earlier recommendation to adopt the benchmark (OpenAI)

OpenAI found widespread task issues in SWE-Bench Pro. Approximately 30% of tasks are broken. OpenAI retracted its earlier recommendation to adopt the benchmark.