Anthropic released Claude Opus 5.5 on 22 September, the first model in a new 5.5 family, and priced it below its predecessor. Input tokens cost $4 per million against Opus 5’s $5, output $20 against $25, and cache reads fall from $0.50 to $0.20.
The headline claim is a 40% cost reduction on typical workloads at, larger than the 20% cut in per-token price, because Anthropic says the model also gets through tasks using fewer tokens.
The more useful part of the announcement is what it concedes. Anthropic’s own comparison table shows GPT-6 Astra ahead of Opus 5.5 on two measures: AutomationBench, 41.4% against 40.0%, and Terminal-Bench-Science, 64.6% against 58.7%. The company also writes that at current capability levels “benchmark margins have become a less reliable guide to real-world differences”, and that in its own use the gap between Opus 5.5 and Claude Fable 5.1 is narrower than the published scores suggest. That is an unusual thing for a vendor to print directly beneath its own benchmark table.
Opus 5.5 was tested before release by outside evaluators including Frontier Design and METR. It is available now on the Claude Platform, AWS, Google Cloud and Azure.