From the team

Both Labs Asked for a Say in What Their Independent Reviewers Publish

Reuters: Anthropic and OpenAI sought control over what outside evaluators publish, and an independent run of a merge-quality coding benchmark tops out at 52%.