Claude Opus 4 and Sonnet 4 launch; Opus 4 deployed under ASL-3 safeguards

Anthropic/Claude models

Anthropic released Claude Opus 4 and Claude Sonnet 4, hybrid models that answer quickly or reason at length and can now use tools like web search while thinking. Opus 4, pitched for long-running coding and agent tasks, shipped with stricter ASL-3 safety protections; both support extended thinking with parallel tool execution.

• Opus 4: best for long-running coding and complex tasks; 72.5% on SWE-bench.
• Sonnet 4: balance of capability and efficiency; 72.7% on SWE-bench; 65% less likely to take shortcuts.
• Both support extended thinking with tool use (web search, code execution); parallel tool execution.
• Opus 4 ships with stricter ASL-3 safety protections; available via API, Claude.ai, Bedrock, Vertex.