Opus 5.5 also led the Artificial Analysis Coding Agent Index, scoring 66 at maximum effort in Claude Code, compared with 60 for Opus 5 and 62 for Claude Fable 5.1. That improvement came at a higher cost: it used $13.04 per task, up 21% from Opus 5’s $10.79, as greater token use outweighed lower token prices.
Released Sept. 22, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, both down 20% from Opus 5. Anthropic’s tests at default settings showed a 40% reduction in cost on typical workloads, distinct from the maximum effort settings used in the coding evaluation. The model also generates output more than 30% faster than Opus 5.