Separating signal from noise in coding evaluations
Via OpenAI Blog
Wednesday, Jul 8, 2026 · 1:00PM
Summary
A new analysis from OpenAI reveals issues in SWE-Bench Pro, a popular coding benchmark, raising concerns about reliability and accuracy in evaluating AI models.
