ainewsblitz.com

Breaking

OpenAI Retracts SWE-Bench Pro Endorsement, Citing ~70% Noise Ceiling

  • Software Dev & Coding
  • Research & Papers
  • AI Agents

OpenAI announced on July 8, 2026 that after auditing SWE-Bench Pro, one of the most widely used benchmarks for evaluating AI coding capability, it found the eval no longer reliably measures frontier coding capability and is retracting its recommendation to the research community. The company said the evaluation is saturated at a "~70% noise ceiling," making it less reflective of true capability.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 4,722 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year