Anthropic announced the release of Claude Opus 5.5, describing it as more capable and cheaper than the previous Opus version.

Performance and benchmark results

The company states that the new model operates at the level of Fable 5.1 across most evaluated tasks, while reducing usage costs by 40% compared with Opus 5.

In published comparisons, Opus 5.5 shows improvements in programming, document handling and desktop automation relative to earlier models.

Specifically, in Terminal-Bench 4.0 Opus 5.5 scored 66.4%, compared with 57.9% achieved by GPT-6 Astra on the same benchmark.

Anthropic also reported that Astra retained advantages in scientific problem solving and in automating business processes, according to the published tests.

Subscription features and developer pricing

For paying users, Anthropic increased the duration of concurrent usage by extending five-hour limits and added a single reset that can be stored for later use.

For developers, the company reduced prices for input and output tokens by 20%, and cut the cost of cache reads by 60%, aiming to lower integration expenses.

Upcoming releases

Anthropic said it plans to publish Sonnet 5.5 and Haiku 5.5 within the coming weeks, expanding the Opus family and related offerings.

The company did not provide specific dates for broader availability or detail additional roadmap milestones in the announcement.