Anthropic Unveils Claude Opus 5.5 With Lower Costs and Stronger Agentic Coding Capabilities
Anthropic has launched Claude Opus 5.5, the first model in its new Claude 5.5 family, introducing improvements in software development, AI agent operations, and complex professional workloads while lowering usage costs compared with its previous Opus model.
The company said Claude Opus 5.5 delivers performance close to Claude Fable 5.1 across a wide range of tasks, while cutting operating costs by approximately 40% on typical workloads compared with Claude Opus 5.
Anthropic described the new model as the most capable version of Opus to date. API pricing has been set at $4 per million input tokens and $20 per million output tokens, down from $5 and $25, respectively, for Opus 5. The cost of reading cached tokens has also been reduced to $0.20 per million tokens, representing a 60% decrease.
The company added that Opus 5.5 generates outputs more than 30% faster than Opus 5, while a new Fast Mode available through Claude Code and the Claude platform can deliver speeds of up to 2.5 times faster.
Focus on Complex Workflows
Claude Opus 5.5 is designed to handle long-running and complex tasks, particularly agentic coding. Anthropic said the model can perform large-scale migrations across extensive codebases, identify software issues, review code, and manage multi-tool workflows for extended periods with limited human intervention.
In Anthropic's internal evaluations, Opus 5.5 achieved 66.4% on Terminal-Bench 4.0, compared with 52.3% for Opus 5 and 57.9% for GPT-6 Astra. It also scored 54.4% on FrontierCode v1.1 and 1,846 points on GDPval-AA v2.1, which evaluates AI performance on real-world knowledge-intensive professional tasks.
Enhanced Safety Measures
Safety also featured prominently in the launch. Anthropic said Claude Opus 5.5 is the first Opus model to introduce safeguards similar to those used with Claude Fable 5.1 for areas including cybersecurity, biological risks, and attempts to replicate model capabilities.
The model underwent external assessments ahead of its release, including evaluations conducted by METR and Frontier Design.
Anthropic said Opus 5.5 incorporates mechanisms that assess actions before they are executed, alongside an open-source isolated runtime environment that security teams can inspect. It also includes a code-review system designed to identify vulnerabilities before code is integrated.
According to the company, the model demonstrated stronger resistance to prompt-injection attacks than Opus 5 across tests involving coding, tool use, web browsing, and computer control.
Anthropic also said Opus 5.5 achieved its strongest performance to date in the company's automated behavioral alignment evaluations, which simulate thousands of scenarios to assess whether the model follows predefined safety boundaries.
Claude Opus 5.5 is currently available to Pro, Max, Team, and Enterprise subscribers through the Claude platform. Developers can also access it through the Claude API and Claude Code, as well as through Amazon Web Services, Google Cloud, and Microsoft Foundry.
Anthropic said it plans to introduce Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks, expanding the new Claude 5.5 model family.