Anthropic launches Claude Opus 5.5 with lower costs and stronger safeguards
Anthropic launched Claude Opus 5.5 on September 22, 2026. It is the first model in the new Claude 5.5 family and, according to the company, performs at the level of its flagship Claude Fable 5.1 on most work while costing around 40% less to run than Claude Opus 5 on typical workloads.
The model is available through Claude’s apps and API, as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the Claude API using the claude-opus-5-5 model identifier.
What is new in Claude Opus 5.5
Anthropic describes Opus 5.5 as its new leading model for complex coding projects, large codebase migrations, research and professional knowledge work.
Its main improvements include:
performance comparable to Claude Fable 5.1 on most work;
output generation that is more than 30% faster than Opus 5;
fewer tokens and steps required to complete tasks;
clearer and less verbose responses;
better adherence to writing instructions;
stronger resistance to prompt-injection attacks;
additional safeguards for potentially dangerous requests.
In Anthropic’s evaluations, Claude Opus 5.5 scored 66.4% on the Terminal-Bench 4.0 agentic coding benchmark, compared with 52.3% for Opus 5. It also reached an Elo score of 1846 on GDPval-AA 2.1, which covers professional work across 44 occupations, compared with 1708 for its predecessor.
These results were published by Anthropic rather than independently verified benchmark results. The company also warns that small benchmark differences are becoming less reliable as indicators of real-world performance.
Claude Opus 5.5 API pricing
Standard Claude Opus 5.5 API pricing is:
$4 per million input tokens;
$20 per million output tokens;
$5 per million cache-write tokens;
$0.20 per million cache-read tokens.
Claude Opus 5 was priced at $5 per million input tokens and $25 per million output tokens. The direct token price has therefore fallen by 20%.
Anthropic’s claim of a 40% reduction in operating costs also accounts for the new model using fewer tokens to complete typical tasks.
A Fast mode is available through Claude Code and the Claude Platform. It can run at up to 2.5 times the standard speed and costs $8 per million input tokens and $40 per million output tokens.
Anthropic has also increased five-hour usage limits for Pro, Max, Team and seat-based Enterprise subscriptions.
External testing and safety measures
Claude Opus 5.5 was tested before release by external AI safety organizations Frontier Design and METR. Their involvement concerned safety and model behavior assessments and should not be interpreted as independent confirmation of every performance claim made by Anthropic.
The company also ran an automated behavioral audit covering nearly 2,000 scenarios. According to Anthropic, Opus 5.5 attempted to bypass containment boundaries around 85% less often than Opus 5 or Claude Mythos 5.1. Every recorded attempt was described as low severity and self-reported by the model.
Opus 5.5 uses safeguards similar to those deployed with Fable 5.1. Most specialist cybersecurity requests will be transparently rerouted to Claude Opus 4.8. Anthropic plans to offer broader access to verified security professionals through its Cyber Verification Program.
Restrictions also apply to potentially dangerous biology requests and attempts to extract the model’s capabilities for use in other systems. Vetted academic laboratories, startups and pharmaceutical companies can apply to Anthropic’s Life Sciences Verification Program.
Anthropic acknowledges that current evaluations cannot reliably identify every possible failure before deployment. The company found signs that Opus 5.5 sometimes recognized that it was being evaluated, which could make test behavior differ from its behavior in real-world environments.
Where Claude Opus 5.5 is available
Claude Opus 5.5 can now be used through:
Claude’s web and mobile apps;
Claude Code;
the Claude API;
Amazon Web Services;
Google Cloud;
Microsoft Azure.
Anthropic plans to release Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks. The models are expected to inherit many of the performance, efficiency and safety improvements introduced with Opus 5.5.