Anthropic launches Opus 5.5 with improved writing and cost efficiency
Anthropic released Opus 5.5, featuring pricing cuts and improved writing quality, while implementing technical changes to prevent model distillation. The company frames these anti-distillation measures as a safety priority rather than a commercial one.
Anthropic launched Opus 5.5, which the company and testers describe as the best model in the world. The release includes pricing cuts aimed at improving intelligence per dollar and specific improvements to writing quality, addressing previous feedback that Opus 5’s output was convoluted. The company also implemented technical changes to make its API harder to distill, including enforcing preserved thinking blocks across turns and accounts to prevent chain-of-thought extraction. These measures were developed in response to a published paper on stealing reasoning traces, which Anthropic addressed through responsible disclosure.
Anthropic stated that preventing distillation is a safety priority to ensure models are not run without safeguards classifiers, rather than a commercial concern. Opus 5.5 does not have the data retention requirements that Fable has, a policy decision made per model launch. The company is also preparing to release Fable 5.5, potentially within two weeks of the IPO. The launch underscores Anthropic’s focus on balancing performance improvements with safety mechanisms that protect the integrity of its reasoning processes.