Artificial intelligence / news brief
Artificial Analysis rebuilds its coding-agent index around mergeability
- Event
- Published
Artificial Analysis released Coding Agent Index v2.0, replacing DeepSWE with Cognition's 100-task FrontierCode benchmark and revising repository-Q&A grading. The index now covers 290 tasks and scores model-plus-harness combinations; its private FrontierCode set limits outside inspection, and v2 scores are not comparable with older versions.
Why it matters
The update shifts a visible agent leaderboard toward whether code changes would be mergeable, while exposing how strongly benchmark design can move rankings.
Sources
- Coding Agent Index v2.0 Methodology (external site)Artificial Analysis · primary source ·
- Artificial Analysis Ranks Claude Opus 5.5 First in Tougher Coding Agent Index (external site)AlphaSignal · coverage ·