Benchmarks

SWE-bench Verified hit its ceiling. That is useful information.

OpenAI says it has stopped reporting SWE-bench Verified for frontier coding models. Builders should read that less as drama and more as a reminder: benchmark confidence expires.