Anthropic has released Claude Opus 5.5, the latest version of its most advanced Claude model tier, with improvements focused on coding, knowledge work, pricing, speed, and safety evaluation.
Opus is the highest-capability model in Anthropic’s three-tier Claude lineup, positioned above Sonnet and Haiku. Anthropic said Opus 5.5 sets new performance levels in coding and knowledge tasks, outperforming the larger Fable model across several benchmarks and completing some informal tasks that Fable could not finish.
Lower Costs and Improved Performance
Anthropic has also reduced the cost of using its top-tier model. Output tokens for Opus 5.5 are priced at $20 per million tokens, compared with $25 per million tokens for the previous version.
The company said the new model is also faster to run because it requires less computing power. Anthropic has made additional changes to how the model communicates, making responses less dependent on technical language and placing important information earlier in answers.
Expansion of the Claude Model Family
The launch comes two months after the release of Claude Opus 5. Anthropic said Sonnet 5.5 and Haiku 5.5, the other models in its lineup, are expected to arrive in the coming weeks with similar improvements.
Safety Measures for Advanced AI Systems
Anthropic said Opus 5.5 has safeguards similar to those used for its Fable model because of its capabilities in areas such as biology and cybersecurity. These restrictions limit certain uses, including helping discover software vulnerabilities in compiled programs and developing biological threats.
The company has continued expanding its safety work, including updates to its approach for handling advanced biological capabilities through its biology safety safeguards.
The release is Anthropic’s first major model launch since CEO Dario Amodei outlined a more cautious approach to advancing frontier AI systems. In his statement on pacing frontier AI development, Amodei said safety measures need time to develop alongside increasingly capable AI systems.
Anthropic said Opus 5.5 underwent alignment testing and pre-release evaluations with external organisations, including METR and Frontier Design. The company also said it is preparing additional security and monitoring systems for future models.
The company expects to provide more information about its safety infrastructure and evaluation processes as those efforts continue.
