Anthropic releases Opus 5.5 with lower prices and Fable-level performance

Anthropic released Opus 5.5 on Tuesday, and the pitch is simple enough: better model, lower bill. The company says it has set a new state of the art in coding and knowledge work, while also coming in below the larger Fable model on many benchmarks. In some informal tasks, Anthropic says Opus 5.5 did things Fable could not finish.

The pricing cut is the part most customers will notice first. Output tokens now cost $20 per million, down from $25 in the previous model. Other pricing metrics got similar reductions, and Anthropic says the model is faster to run too, which points to less compute needed to serve it.

There’s also a change in how the model talks. Anthropic says Opus 5.5 is less likely to use jargon and more likely to lead with the important part of the message. That sounds minor until you’ve spent time fighting a model that buries the answer three paragraphs deep.

The release comes only two months after Opus 5, which arrived on July 24. Anthropic says Sonnet 5.5 and Haiku 5.5 are coming “in the coming weeks,” with similar gains. The company also says Opus 5.5 matches Mythos on biology and cybersecurity capabilities, so it falls under the same safeguards as Fable, including limits on using the models to find exploits in compiled programs or develop recognizable biological weapons.

This is Anthropic’s first model release since Dario Amodei argued for pacing frontier AI development so safety work can keep up. The company says Opus 5.5’s training was broadly similar to earlier models, with alignment testing and outside evaluation from groups including METR and Frontier Design. It also says more advanced training and evaluation systems are already being prepared for future models, along with better security and monitoring.