Anthropic unveils Opus 5.5, a faster, cheaper AI model that outperforms Fable

Anthropic released Opus 5.5 on Tuesday, positioning it as the strongest-performing model the firm has tested to date, especially in coding and knowledge-work tasks. The new system outpaces the larger Fable model on many benchmarks and succeeds in informal tasks that previously stumped Fable. Pricing for output tokens drops to $20 per mTok, a reduction from the $25 rate of its predecessor, and the model also runs faster due to lower compute demands.

Communication has been tweaked to avoid jargon and place crucial details at the beginning of responses. The rollout follows the July 24 launch of Opus 5, with Sonnet 5.5 and Haiku 5.5 slated for release in the coming weeks. Opus 5.5 inherits the same safety constraints as Fable, limiting its use for exploit discovery or biological weapon development, and its safety testing involved external reviewers like METR and Frontier Design.