Updates·September 23, 2026, 21:09

Claude Opus 5.5: 40 percent cheaper and faster for code

AI-generated and checked against the sources listed below.

Anthropic is launching Claude Opus 5.5, which according to the company is more than 30 percent faster, stronger at programming and up to 40 percent cheaper to use than its predecessor Opus 5.

What's new

On September 22, 2026, Anthropic launched Claude Opus 5.5, the newest and most powerful version in the company's AI family. The model replaces Opus 5 and is available via Anthropic's own platform as well as on Amazon Web Services, Google Cloud and Microsoft Azure.

What's clever

Opus 5.5 responds more than 30 percent faster than its predecessor, and there is now also a "fast mode," which according to Anthropic can be up to 2.5 times as fast. The company is also raising the 5-hour usage limit that many Pro, Max and Team subscribers know by about 20 percent, and users can now save a reset of the limit for later use. Anthropic also says the model writes more plainly and less "artificially" than earlier Claude versions, with fewer technical terms and the most important point first.

Cheaper or better?

For developers using the model via the API (the technical access route that apps and companies use), the price has been cut: input now costs $4 per million tokens (pieces of text) versus $5 before, and output costs $20 versus $25 previously, a 20 percent drop. Reuse of earlier responses, called caching, has become up to 60 percent cheaper. Overall, Anthropic expects a typical task to cost about 40 percent less than with Opus 5. Independent tests also show that Opus 5.5 beats OpenAI's GPT-6 Astra on the coding test FrontierCode at roughly one-fifth of the price. The price of the regular Pro subscription has not been reported to have changed with the launch.

What it's good at

The model is especially strong at programming and complex work tasks. On Terminal-Bench 4.0, which measures the ability to solve tasks in a command line, it scores 66.4 percent, and on the coding test FrontierCode 54.4 percent. According to Anthropic, one tester used Opus 5.5 to rewrite a program of 680,000 lines of code in less than a day. The company also highlights improved safety: attempts to bypass the model's guardrails, so-called jailbreaks, have dropped significantly compared with earlier versions.