Mercor

Mercor launched APEX-SWE, a benchmark created in collaboration with Cognition that measures whether frontier models can handle real production software engineering work such as integrations and observability debugging. At release, the top score was GPT-5.3 Codex at 41.5% Pass@1.