AIClaude Opus 5.5’s reported coding benchmark gains and higher cost per task
Artificial Analysis reports its highest Coding Agent Index score yet: 66 for Opus 5.5 at max effort in Claude Code. But cost per task rose 21% over Opus 5, despite lower token prices.
TLDR
Artificial Analysis says Opus 5.5 scored 66 on its Coding Agent Index at max effort in Claude Code, up from 60 for Opus 5 and 62 for Claude Fable 5.1. It improved across all three evaluations, with the largest gain on Terminal-Bench 4.0: 63.1%, up from 54.5% for Opus 5.
The evaluator reports that input/output pricing fell 20% to $4/$20 per million tokens, but Opus 5.5 used substantially more tokens. Cost per task reached $13.04, versus $10.79 for Opus 5—the highest in its comparison.
Separately, a user says rebuilding a three.js version of “The Peach Blossom Spring” with Opus 5.5 produced much better results. They emphasize that it took repeated refinement, not one prompt, and that they had the model search for free 3D models to avoid building them from scratch.
