Claude Opus 5.5 tops the Coding Agent Index, at higher cost per task

Claude Opus 5.5 costs more per coding task than Opus 5 because it uses far more tokens, and the extra usage outweighs a lower price per token. Anthropic cut Opus pricing to $4 per million input tokens and $20 per million output tokens, down from $5 and $25, and cut cache reads to $0.20 from $0.50. Yet Artificial Analysis measured an average cost of $13.04 per task on its Coding Agent Index, 21% above Opus 5's $10.79. Opus 5.5 used about 15.6 million tokens per task against 11.4 million for Opus 5, including about 2.4 times as many output tokens, which are the most expensive kind.

The extra spending buys the highest score Artificial Analysis has measured on the index. The Coding Agent Index averages three evaluations with equal weight, and here each was run in Claude Code at max effort. Opus 5.5 scores 66, against 60 for Opus 5 and 62 for Claude Fable 5.1. It improved on all three: Terminal-Bench 4.0 rose to 63.1% from 54.5%, the largest gain, DeepSWE v1.1 to 68.4% from 62.5%, and SWE-Atlas-QnA to 66.4% from 62.1%. Artificial Analysis says no lower-cost model in its comparison matches that score, so Opus 5.5 extends the cost-versus-score frontier at its expensive end rather than making top coding performance cheaper.

The result shows that a lower per-token price does not guarantee a cheaper task: cost per task also depends on how many tokens the model spends. Artificial Analysis found the same pattern on its broader Intelligence Index, where Opus 5.5's cost per task stayed close to Opus 5's despite the price cut. These figures cover max effort in Claude Code only.