Anthropic has launched Opus 5.5, an advanced AI model that significantly enhances performance in coding and knowledge tasks while lowering costs compared to its predecessor. As the most premium tier in Anthropic’s Claude model lineup, which also includes Sonnet and Haiku, Opus 5.5 outperforms the larger Fable model in several benchmarks and succeeds at tasks where Fable struggled. The update reduces output token prices from $25 to $20 per million tokens and offers faster processing by lowering computational demands.

The new Opus 5.5 model incorporates improvements in communication style, favoring clarity by placing key information early and avoiding technical jargon, which enhances user interaction. This release follows closely after Opus 5 debuted in July 2026, with Sonnet 5.5 and Haiku 5.5 expected soon, promising similar gains. Anthropic has also aligned Opus 5.5’s safety protocols with those used for the Fable model, applying strict restrictions to prevent the AI from being used for harmful purposes, such as cybersecurity exploits or biological weapon development.

The upgrade comes during a period of cautious advancement in AI development, championed by Anthropic CEO Dario Amodei, who advocates pacing AI capability growth to better manage associated risks. Amodei stresses that slowing progress allows more time to improve risk prevention measures and aligns with the company’s commitment to balancing innovation with safety. Anthropic’s approach involves rigorous pre-release testing, including external evaluations from groups like METR and Frontier Design, to ensure responsible model deployment.

Anthropic also highlights ongoing efforts to enhance its security, monitoring, and alignment systems as AI capabilities advance. The company foresees a more prominent role for public policy in governing AI safety and is building infrastructure to support these regulatory frameworks. While additional details on these initiatives will be shared in the future, the release of Opus 5.5 marks a notable step forward in delivering powerful, safer, and more affordable AI tools for knowledge work and coding domains.