Positioning and release date
Claude Opus 4 is a model in Anthropic's Claude 4 series, released on May 22, 2025. At launch, it was positioned for complex programming, deep reasoning, and long-running agent tasks. (anthropic.com)
Capabilities and uses
- Supports both quick responses and extended thinking, and can be used for complex problem-solving, writing and modifying code, research, and writing.
- Supports parallel tool calls; when developers provide local file access, the model can save task-related information by reading and writing files. At launch, using tools during extended thinking was a beta feature.
- The officially reported SWE-bench Verified score was 72.5%; this result was obtained under the evaluation settings described by Anthropic and does not represent the success rate for all real-world tasks. (anthropic.com)
Limitations and availability
The model may generate incorrect or unreliable content; code and important conclusions should be verified by a human. Tool calls, file access, and memory capabilities depend on application configuration and authorization and cannot be assumed to be available. At launch, Anthropic stated that the context window was 200K tokens. (anthropic.com)
The model was initially available through the Claude API and other platforms; however, Anthropic retired this model ID from the Claude API on June 15, 2026. Availability on other platforms should be checked against the respective platforms' current information. (platform.claude.com)
Pricing
The historical standard API pricing in the launch announcement was USD 15 per million input tokens and USD 75 per million output tokens. This is historical pricing and does not mean that the model remains accessible through the Claude API after retirement; different terms and prices may apply to caching, batch processing, and other platforms. Official launch and pricing announcement, accessed November 18, 2026. (anthropic.com)