Claude Sonnet 5.5 ranks second on Intelligence Index
Anthropic's Claude Sonnet 5.5 scores 56 on Artificial Analysis's Intelligence Index, 2 points behind Opus 5.5 at unchanged pricing, though record token use raises its per-task cost by about 50%.
Anthropic has released Claude Sonnet 5.5. In independent testing by Artificial Analysis, the model scored 56 on the firm's Intelligence Index at max effort. That is 18 points above Sonnet 5 and 2 points behind Opus 5.5 (max), which puts Sonnet 5.5 at #2 overall.
The biggest gain came in agentic terminal use. On Terminal-Bench 4.0, Sonnet 5.5 scored 64%, up 50 points from Sonnet 5 and ahead of the 60% posted by both Opus 5.5 and GPT-6 Astra. It roughly matched Opus 5.5 on the AA-Briefcase, GDPval-AA and AutomationBench-AA knowledge-work benchmarks. It still trailed Opus on factual accuracy in AA-Omniscience (54% vs 66%), although its hallucination rate was lower (47% vs 59%). It also scored about 6 points lower than Opus on Humanity's Last Exam and SciCode.
Those results took a lot of tokens. At max effort, Sonnet 5.5 used about 193k output tokens per Intelligence Index task. That is the most Artificial Analysis has measured, about 60% more than Opus 5.5 or Sonnet 5 and roughly 7x GPT-6 Astra. List pricing is unchanged at $2/$10 per million input/output tokens, but cost per task rises to $7.60, about 50% more than Sonnet 5. On intelligence versus cost per task, the model sits off the Pareto frontier. Its high-effort setting is its most competitive, landing narrowly behind GPT-6 Sol at about the same cost.
The context window stays at 1 million tokens, and the model offers five effort settings.
The results are provisional. Artificial Analysis tested a pre-release deployment that had a bug affecting structured outputs. Anthropic says the bug is fixed for the public release, and Artificial Analysis plans to re-run the affected evaluations.