Artificial Analysis publishes Coding Agent Index comparing agent costs and benchmarks

Artificial Analysis has published a Coding Agent Index that compares coding agents on per-task cost and benchmark results, linking to an interactive index and a methodology page.

In the index's first reported comparison, Artificial Analysis says Claude Sonnet 5.5 scores 68 at max effort and 63 at xhigh effort in Claude Code. Per task, xhigh costs $3.33 against $14.19 at max, a 77% cut for 5.5 fewer Index points. Max generates 4.8 times as many output tokens and takes 3.4 times as many turns. Artificial Analysis says max improves all three benchmark scores, with the largest gain on Terminal-Bench 4.0, rising from 58.1% to 66.2%.