Claude Opus 5.5
3 hours ago
- Claude Opus 5.5 is a new model that matches Claude Fable 5.1's performance at 40% lower cost than Opus 5.
- It achieved top scores on internal alignment tests and was evaluated by external groups like Frontier Design and METR.
- Performance improvements include completing a 680,000-line code migration in under a day and cutting load times across 39 of 40 web app pages.
- Safety is enhanced with strong alignment, reduced misaligned behavior, and safeguards similar to Fable 5.1 for cybersecurity and biology.
- Cost and speed are improved: input/output tokens are $4/$20 per million (20% less), cache reads $0.20 per million (60% less), and output generation is over 30% faster.
- Communication is clearer and more natural, with important information up front and easier-to-follow writing.
- Upcoming releases include Claude Sonnet 5.5 and Haiku 5.5 with similar improvements.
- Benchmarks show leadership in agentic coding, computer use, and knowledge work, though real-world gaps with Fable 5.1 are narrower.
- Pacing the frontier includes safety practices for current models and preparation for more advanced future models.
- Antidistillation safeguards like preserved thinking prevent extraction of reasoning, and the model is available on AWS, Google Cloud, and Azure with zero data retention.