Anthropic has released Opus 5.5, its most capable Claude model tier, with the company claiming leading results in coding and knowledge work benchmarks. Anthropic also said the model outperformed the larger Fable model on many tests and completed informal tasks that Fable could not.
Opus 5.5 is cheaper and faster than its predecessor. Output now costs $20 per million tokens, down from $25, while other pricing metrics have also declined. The company said the model requires less computing power to run.
The update also changes the model’s communication style. It is less likely to use jargon and more likely to place key information at the beginning of its responses.
Released two months after Opus 5 on July 24, Opus 5.5 will be followed by Sonnet 5.5 and Haiku 5.5 “in the coming weeks,” according to Anthropic.
Anthropic said Opus 5.5 has biology and cybersecurity capabilities comparable to Mythos, placing it under the same safeguards as Fable. These controls restrict certain uses, including discovering exploits in compiled software and developing recognizable biological weapons.
The release is the first since CEO Dario Amodei called for a slower pace of AI capability development to give safety and alignment work more time to advance. Anthropic said Opus 5.5 underwent alignment testing and external pre-release evaluations by METR and Frontier Design, while stronger security and monitoring systems are being developed for future models.
Source: TechCrunch
