Anthropic Unveils Opus 5.5 With Cheaper Pricing and Fable-Level Performance

Anthropic releases Opus 5.5, a new version of its most capable and expensive Claude model that sets a state-of-the-art mark in coding and knowledge work performance. The company says the model outpaces the larger Fable model on many benchmarks and completes informal tasks that Fable fails to finish. Output tokens cost $20 per million, down from $25 for the previous version, and the model runs faster, reflecting lower compute requirements. Anthropic plans to release Sonnet 5.5 and Haiku 5.5 in the coming weeks with similar improvements.

The new model also changes how it communicates, using less jargon and placing important information at the start of its responses. Because Opus 5.5 matches Mythos in biology and cybersecurity capabilities, its release falls under the same safeguards as Fable, limiting its use for finding exploits in compiled programs or developing recognizable biological weapons. The launch arrives just two months after Opus 5 and marks Anthropic's first release since CEO Dario Amodei embraces calls to pace the frontier, deliberately slowing capability gains so risk prevention can keep up.

Safety training for Opus 5.5 remains broadly similar to its predecessors, with alignment testing and pre-release evaluations by outside organizations such as METR and Frontier Design. Anthropic says it is preparing more advanced training and evaluation systems for future models, including improved security and monitoring. The company also argues that public policy should play a larger role in ensuring that increasingly capable AI systems remain reliable for the people who depend on them.

Read More at the original source →