Anthropic launches Claude Opus 5.5 with new token pricing

Anthropic says Claude Opus 5.5 is the first model in its new Claude 5.5 family, and the company is pitching it as both cheaper and more capable than Opus 5. On typical workloads, it should cost 40% less to run. Input tokens are priced at $4 per million, output at $20, and cache reads at $0.20 per million.

The model arrives with a lot of emphasis on safety. Anthropic says it was tested by external evaluators including Frontier Design and METR before release, and that its automated behavioral audit is the strongest-performing model the company has tested so far. It also ships with the safeguards reserved for Anthropic’s most capable models.

Performance is the headline, though. Anthropic says Opus 5.5 is now its leading model, with early testers reporting big gains on long, messy work. One tester used it for a 680,000-line code migration in under a day. Another saw it cut load times across every page of a web app successfully 39 times out of 40. A different group had several Claude models build a game from one prompt, and Opus 5.5 came out on top for graphics and polish.

The company is also making a point about how the model communicates. Early testers said it wrote more clearly, put the important stuff first, and was easier to check over long sessions. Anthropic says that matters for safety too, since cleaner output is easier for humans to inspect. Claude Sonnet 5.5 and Claude Haiku 5.5 are due in the coming weeks, and Anthropic says they’ll share many of the same improvements.

There’s a clear pattern here: the frontier is getting cheaper, but only if you trust the vendor’s own benchmarks, safeguards, and scorekeeping. Anthropic is selling Opus 5.5 as both stronger and more restrained, which is exactly what serious buyers want to hear. The trick is that everybody now claims to be careful while shipping harder-to-ignore models.