“OpenAI has formally retracted its support for the SWE-Bench Pro benchmark, concluding it no longer reliably measures frontier coding capability.”