Anthropic Releases Claude Opus 5.5, Cuts Output Pricing to $20 per Million Tokens

Anthropic released Claude Opus 5.5 on Tuesday, cutting output token pricing to $20 per million tokens from $25 for its predecessor while delivering what the company calls its strongest coding and knowledge-work performance to date.

The release comes just two months after Opus 5 launched on July 24, making this one of Anthropic's fastest model iterations. According to Anthropic, Opus 5.5 outperforms the larger Fable model on multiple benchmarks and completed a number of informal tasks that Fable could not. The company's announcement describes Opus 5.5 as "the strongest-performing model we've tested to date." Anthropic did not disclose specific benchmark scores in its announcement.

Pricing and Performance

Output tokens now cost $20 per million tokens, down from $25 for Opus 5. Anthropic says other pricing metrics saw comparable reductions, though exact input token pricing was not disclosed. The company attributes the lower cost to reduced compute requirements, which also translate into faster inference speeds.

Anthropic also revised how Opus communicates. The model is now less likely to use jargon and more likely to place key information at the start of its responses — a change aimed at knowledge-work use cases where users scan for conclusions first.

Anthropic says Sonnet 5.5 and Haiku 5.5 will follow "in the coming weeks" with similar performance improvements, extending the update across its full model lineup.

Safety Context

Opus 5.5 arrives as the first model release since CEO Dario Amodei publicly committed to slowing capability progress to let alignment work catch up. "I have become convinced that fully addressing the risks requires even more prudence," Amodei wrote earlier this month, "not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up."

Anthropic says Opus 5.5's biology and cybersecurity capabilities are comparable to its Mythos model, triggering the same safeguards applied to that release. Those restrictions limit the model's use in discovering exploits in compiled software and developing biological threats, according to the company.

Safety training for Opus 5.5 followed a process similar to prior releases, including alignment testing and pre-release evaluation by outside groups METR and Frontier Design. Anthropic says it is building more advanced training and monitoring infrastructure for future models. "As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe," the company wrote in its announcement, adding that it plans to share more details on these efforts soon.

What This Means

The pricing cut and speed improvements suggest Anthropic found meaningful compute efficiencies in Opus 5.5's architecture rather than simply scaling up parameters. That's notable given Amodei's stated commitment to decelerate capability gains — this release reads more like an efficiency and usability update than a raw capability leap, consistent with a company trying to prove that safety-paced development doesn't have to mean stagnant products. The two-month cadence since Opus 5 also shows Anthropic can still ship rapidly within its new stated constraints, even as it emphasizes building out policy and monitoring infrastructure it says will govern more capable future models.