TodayToday / Industry / StoryOpenAI Scraps SWE-Bench Pro Over 30% Broken Tasks [2026]ShareWed 8 Jul · 1:00 PM, updated 11 Jul · 7:00 AMFirst seen Wed 8 Jul · 1:00 PMUpdated Sat 11 Jul · 7:00 AMEmerging, taperingOpenAI · IndustryReported by OpenAI and tech-insider.org2 sources ↓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.SourcesOldestNewestOpenAIWed 8 Jul · 1:00 PMRead ↗Separating signal from noise in coding evaluationstech-insider.orgSat 11 Jul · 7:00 AMRead ↗OpenAI Scraps SWE-Bench Pro Over 30% Broken Tasks [2026]More on OpenAIOpenAI launches GPT-Live-1 for ChatGPT VoiceWed 8 Jul · 18 outletsApple sues OpenAI alleging trade secret theft, says scheme was 'at every level'Fri 10 Jul · 17 outlets