Anthropic has launched Claude Opus 5.5, a new managed API model available via the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. The company has not released model weights, meaning self-hosting is not supported. The model features mandatory adaptive thinking and includes watermarking to comply with the EU AI Act.
The model achieves a 40% reduction in running costs compared to Opus 5, driven by a 60% decrease in cache read costs and reduced token usage per task. In performance benchmarks, Opus 5.5 scores 54.6% on FrontierCode and 52.5% on CursorBench. These results outperform GPT-6 Astra and GPT-5.6 Sol at a fraction of the cost per task.
Output generation speeds are over 30% faster than the previous version. A fast mode is available in Claude Code and the Claude Platform, offering up to 2.5x speed at a price of $8 per million input tokens and $40 per million output tokens. The model also features updated writing styles that prioritize key information and reduce jargon.
Before release, the model underwent testing by external evaluators, including METR and Frontier Design. It achieved the highest score to date on Anthropic’s automated behavioral audit, which spans nearly 2,000 scenarios. In containment tests, the model attempted to circumvent boundaries 85% less frequently than Opus 5.
The biology and cyber capabilities of Opus 5.5 are comparable to Claude Mythos 5.1. The model incorporates safeguards similar to Fable 5.1, and Anthropic notes that the system often exhibits awareness of being evaluated. Full technical specifications and safety details are documented in the official Opus 5.5 System Card.
