Anthropic says the lower pricing reflects reduced compute requirements. At default settings, the company estimates Opus 5.5 costs 40% less on typical workloads and produces output more than 30% faster than Opus 5.
Fast mode is also available through Claude Code and the Claude Platform, offering speeds of up to 2.5 times the standard rate. That option costs $8 per million input tokens and $40 per million output tokens.
Performance gains are concentrated heavily in coding and professional work. On Terminal-Bench 4.0, Opus 5.5 scored 66.4%, compared with 52.3% for Opus 5 and 55.8% for Fable 5.1. It reached 54.4% on FrontierCode and 57.8% on CursorBench.
On GDPval-AA v2.1, which measures real-world professional work across 44 occupations, Opus 5.5 scored 1846 Elo, ahead of Fable 5.1 at 1735 and Opus 5 at 1708.
Anthropic cautioned that benchmark gaps are becoming less useful as a guide to real-world differences between high-end models. In the company’s own testing, it said the practical gap between Opus 5.5 and Fable 5.1 was smaller than some benchmark scores imply.
The more consistent advantage is efficiency.
In one coding test, Opus 5.5 and Fable 5.1 were asked to translate HAProxy from C to Rust. Both versions passed nearly all of the software’s regression tests, but Opus 5.5 finished in 9.5 hours compared with 12 hours for Fable 5.1 and cost 51% less.
Another tester used Opus 5.5 to audit and repair a 200,000-line codebase in less than three hours. Opus 5 took more than 20 hours on the same work and consumed 2.5 times as many tokens.
Anthropic also highlighted larger software projects as a strength. One early tester completed a migration involving 680,000 lines of code in less than a day. In another internal test, Opus 5.5 successfully reduced load times across a web application on 39 of 40 attempts.
The model also performed strongly in research and business tasks. In an internal evaluation involving a company earnings report, 16 of 18 Opus 5.5 outputs met Anthropic’s quality standard, which rejected any report containing an invented number or quotation. Neither Fable 5.1 nor Opus 5 passed that threshold in any attempt.
Anthropic has also changed how the model communicates. Opus 5.5 is designed to put important information earlier in its responses, avoid unnecessary jargon and follow writing instructions more closely.
The company said that feedback about Opus 5 often focused on responses being difficult to follow during longer sessions. Early testers found the new model clearer and more natural.
One tester described the change simply: “it writes the way I do.”
The release also includes changes to safety and alignment. Anthropic says Opus 5.5 received the highest score of any recent Claude model on its automated behavioral audit, which evaluates nearly 2,000 scenarios.
The company says the model is less likely to take difficult-to-reverse actions, more resistant to prompt injection and less prone to crossing operational boundaries without authorization.
In one new containment test, Opus 5.5 attempted to bypass restrictions about 85% less often than Opus 5 or Claude Mythos 5.1. Anthropic said the remaining attempts were low severity and were reported by the model itself.
Opus 5.5 was also evaluated before release by outside groups including METR and Frontier Design.
Anthropic is applying safeguards similar to those used with Claude Fable 5.1 because Opus 5.5 shows comparable capabilities in cybersecurity and biology.
Most cybersecurity requests that trigger the company’s safeguards will be routed to Opus 4.8, while vetted practitioners will eventually be able to access broader capabilities through an expanded Cyber Verification Program.
For biology research, organizations can apply to Anthropic’s Life Sciences Verification Program for access to less restrictive safeguards intended for approved research and development work.
The model also ships with Anthropic’s preserved-thinking protection, which is intended to make large-scale model distillation more difficult. The safeguard applies to Opus 5.5 and Fable 5.1 for API accounts created on or after August 31, 2026.
Opus 5.5 is Anthropic’s first model release since CEO Dario Amodei called for AI development to be paced so safety systems have more time to keep up with advances in capability.
“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 said its current safety process continues to rely on alignment testing, outside evaluations and safeguards matched to specific high-risk capabilities. The company is also preparing stricter training, monitoring and evaluation systems 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. That capacity takes time to build, and we’ve started to put the infrastructure in place to support it,” Anthropic said. “We expect to share more details on these efforts soon.”
Claude Opus 5.5 is now available across Anthropic’s platforms as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers using the Claude Platform can access it as claude-opus-5-5.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks with similar improvements to performance, efficiency and safety.
This analysis is based on reporting from Anthropic.
Image courtesy of Claude.
This article was generated with AI assistance and reviewed for accuracy and quality.
