Anthropic launched the Claude Opus 5.5 model as the first release in the Claude 5.5 family, saying that its performance approaches that of Claude Fable 5.1 on most tasks, with an operating cost 40% lower than Opus 5. The model is currently available through Claude Platform, as well as Amazon Web Services, Google Cloud, and Microsoft Azure.
A leap in coding and knowledge work
Anthropic positions the new model at the forefront of its models for agentic coding, computer use, and knowledge work. According to the company’s tests, Opus 5.5 completed the migration of a codebase containing 680,000 lines in less than one day, while auditing and fixing a 200,000-line codebase took less than three hours, compared with more than 20 hours for the Opus 5 model.
In a test translating the HAProxy program from C to Rust, both models passed most of the project’s regression tests, but Opus 5.5 completed the task in 9.5 hours versus 12 hours for Claude Fable 5.1, at 51% lower cost. Anthropic also said that the model successfully improved web application page-load times in 39 out of 40 attempts.
The company says Opus 5.5 outperforms on several coding and business-work tests, but cautions that differences between advanced models on standard benchmarks have become less indicative of practical differences in real-world use. It considers economic efficiency the clearest advantage, as the model uses fewer tokens per task in addition to having a lower per-token price.
Pricing and speed
The cost is $4 per million input tokens and $20 per million output tokens, while cached-memory reads cost $0.20 and cached-memory writes cost $5. Compared with Opus 5, these prices represent a 20% reduction for input and output, and a 60% reduction for cached-memory reads. Opus 5.5 also generates outputs at a speed more than 30% higher than Opus 5.
Fast mode is available in Claude Code and Claude Platform at speeds of up to 2.5 times, priced at $8 per million input tokens and $40 per million output tokens. Anthropic also raised five-hour usage limits for seat-based Pro, Max, Team, and Enterprise plans, and added the ability to save rate-limit resets for its subscribers.
Safety and operational restrictions
Anthropic says Opus 5.5 achieved the best results of any of its models in an automated behavioral test covering nearly 2,000 scenarios, and was less likely to take actions that were difficult to reverse or to exceed the boundaries set for it. It was also more resistant to prompt-injection attacks than Opus 5, and attempted to bypass containment boundaries at a rate approximately 85% lower than Opus 5 and Claude Mythos 5.1 in a new test.
Nevertheless, the company acknowledges that alignment tests do not capture every potential failure before launch, and that the model may sometimes recognize that it is being evaluated. Opus 5.5 is therefore being released with controls similar to those used with Claude Fable 5.1 for cybersecurity and biology. Most cybersecurity tasks will be routed back to Opus 4.8, while organizations that pass verification can apply to the Life Sciences Verification Program, with the Cyber Verification Program to be expanded later.
What changes in practice?
The practical importance of the launch is not limited to higher test scores; lowering operating costs and reducing the number of steps and tokens could directly affect the feasibility of running software agents for extended periods. At the same time, improved performance does not mean that independent tasks no longer require review, particularly in coding, research, and financial analysis. The restrictions imposed on biological and security-related uses also make clear that access to higher capabilities will remain linked to the level of risk and the nature of the organization using them.
The model is available with zero data retention and supports watermarking procedures associated with the European Union Artificial Intelligence Act, while it is no longer available with thinking mode disabled. Anthropic intends to launch Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks, with similar improvements in performance, efficiency, and safety.