Tag
1 article
OpenAI has found that around 30% of tasks in the popular SWE-Bench Pro benchmark are broken, leading the company to withdraw its endorsement of the test.
