Claude Opus 4
Claude Opus 4 launches as Anthropic's flagship coding and agent model, 72.5% on SWE-bench Verified, deployed under ASL-3 safeguards.
Built for sustained, long-running agentic work. SWE-bench Verified 72.5% and Terminal-bench 43.2% (company-run). Adds extended thinking interleaved with tool use (beta), parallel tool execution and better memory-file use. Anthropic says both Claude 4 models are 65% less likely than Sonnet 3.7 to take shortcut loopholes.
- Date
- Thursday, 22 May 2025
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- $15/$75 per M tokens in/out (2025-05-22)
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-bench Verified | 72.5% Claude Opus 4 | company |
| Terminal-bench | 43.2% Claude Opus 4 | company |
First Claude model released under ASL-3 protections, as a precaution (see anthropic-asl3-activation). Retired from the Claude API on 2026-06-15 (platform release notes). 'World's best coding model' is Anthropic's claim.
Sources
This record was checked against its sources on 6 October 2026. How we check