Artificial Analysis Launches Coding Agent Index: Opus 4.7 and GPT-5.5 Lead, Open Weights Lag Behind
A new benchmark evaluates coding agents as complete systems — model plus harness — across SWE-Bench, Terminal-Bench, and SWE-Atlas, revealing sharp cost and speed differences that matter more than raw accuracy.
Artificial Analysis released its Coding Agent Index on Monday, a benchmark that evaluates not just models but full model-plus-harness combinations on real-world software engineering tasks. As @ArtificialAnlys announced, Anthropic's Opus 4.7 and OpenAI's GPT-5.5 currently lead the rankings, while open-weight models remain "competitive but trail" — a polite way of saying the gap persists where it matters most.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.