Claude Opus 5.5 Tops Intelligence Index at 58 — Fast, Verbose, and Costly

Anthropic’s Claude Opus 5.5, released September 22, 2026, lands near the top of Artificial Analysis’s Intelligence Index with a score of 58, more than double the 25 median for comparable reasoning models. It is a multimodal reasoning model that accepts text and images, outputs text, and ships with a 1M-token context window. The model uses extended chain-of-thought reasoning and is served through five API providers, though it remains proprietary with no disclosed weights or parameter count.

That intelligence carries a premium. Input runs $4 per million tokens and output $20 per million—roughly double the field medians of $2 and $10—yielding a blended rate of about $2.94 per million on a 7:2:1 cache/input/output mix. Cost is compounded by verbosity: the model burned through 260M output tokens completing the benchmark suite versus a median of 88M, pushing the tab for a single Intelligence Index evaluation to $8,708.

The takeaway is a familiar frontier-model tradeoff. Opus 5.5 buys top-tier reasoning performance, but its combination of above-average token pricing and heavy output means real-world costs can climb faster than the headline per-token rates suggest, making it best suited to high-value tasks where accuracy justifies the spend.