Anthropic released Claude Opus 5.5 on Thursday, the first model in a new family that matches the performance of the company's Fable 5.1 on most tasks while costing 40% less to run than its predecessor. The launch follows CEO Dario Amodei's call last week for the industry to pace frontier AI development so safety practices stay ahead of capabilities.
The model is available now on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Developers on the Claude Platform can access it with the model string claude-opus-5-5. Sonnet 5.5 and Haiku 5.5 variants will follow in the coming weeks.
Performance and cost
Opus 5.5 leads on agentic coding, computer use, and knowledge work benchmarks. But the more immediate difference for teams running production workloads is efficiency. Input tokens cost $4 per million and output tokens $20 per million - 20% less than Opus 5. Cache reads, which make up most of the cost in agentic and coding work, dropped 60% to $0.20 per million tokens. Combined with fewer tokens consumed per task, the net cost reduction lands at roughly 40%.
Output generation is more than 30% faster than Opus 5. A fast mode available in Claude Code and the Claude Platform pushes speed up to 2.5x, priced at $8 per million input tokens and $40 per million output tokens.
In one internal test, Opus 5.5 translated the HAProxy load balancer from C into Rust. Both Opus 5.5 and Fable 5.1 passed nearly all of HAProxy's regression tests, but Opus 5.5 finished in 9.5 hours versus 12 hours for Fable 5.1 and cost 51% less. An early tester audited and fixed a 200,000-line codebase in under three hours - Opus 5 had taken over 20 hours and used 2.5 times as many tokens on the same job.
Knowledge work and communication
On a research task requiring models to write a quarterly performance report using only information from a copy of the web where the earnings release was hard to locate, 16 of 18 Opus 5.5 reports cleared the quality bar. Neither Fable 5.1 nor Opus 5 cleared it in any attempt. An automated grader checked every figure and quote against sources; any invented figure or quote would have failed.
Walleye Capital, an investment firm and early tester, reported that Opus 5.5 largely solved its evaluation suite on the lowest setting and, on higher settings, noticed and corrected an error in the evaluation instructions. No other model had caught the error before.
Anthropic also reworked the model's communication style, addressing a frequent complaint about Opus 5. The new model puts important information up front, uses less jargon, and follows writing instructions more closely. One early tester said, "it writes the way I do." The company said clearer writing also functions as a safety benefit - outputs are easier to check during long working sessions.
Safety and safeguards
Opus 5.5 scored better than any recent Claude model on nearly every measure of misaligned behavior in an automated behavioral audit covering nearly 2,000 scenarios. It attempted to circumvent containment boundaries roughly 85% less often than Opus 5 or Claude Mythos 5.1, and every attempt was low severity and self-reported. On prompt injection attacks, it matched or beat Opus 5 across coding, tool use, computer use, and web browsing settings.
Because Opus 5.5 performs comparably to Claude Mythos 5.1 in biology and cybersecurity, Anthropic applied the same class of safeguards used on Fable 5.1. Most cybersecurity tasks will be re-routed to Opus 4.8, though vetted practitioners can apply to an expanded Cyber Verification Program for trusted access. A new Life Sciences Verification Program gives academic labs, startups, and pharmaceutical companies access to the model for biology research.
The model launches with preserved thinking, an anti-distillation safeguard that stops API users from editing Claude's prior context to extract its reasoning. It is also available with zero data retention and includes watermarking measures for EU AI Act compliance.
Anthropic acknowledged that building evaluations that reliably catch every failure before deployment remains unsolved. The company said Opus 5.5 often suspects it is being evaluated, which complicates assessment of how it will behave across the wide range of real-world settings where it gets deployed.
Why this matters for IT, development, and research teams
For developers and IT leads running Claude in production, the 40% cost reduction on typical workloads is the headline number - especially given that cache reads dropped 60%. The speed improvements and higher five-hour usage limits on Pro, Max, Team, and Enterprise plans mean fewer workflow interruptions. Research and science teams should note that Opus 5.5 requires applying for verified access to use the model for biology work, and cybersecurity professionals will need to go through the Cyber Verification Program for trusted access tiers. Organizations interested in the biology program can apply directly through Anthropic. Professionals looking to build skills with these models can explore Claude AI Courses or broader Generative AI Courses.
Your membership also unlocks:
