Anthropic has released its next-generation AI models, Claude Opus 4 and Claude Sonnet 4, introducing new safety measures designed to prevent the misuse of these systems in developing chemical, biological, radiological, or nuclear (CBRN) weapons.
Claude Opus 4 is Anthropic’s most advanced model, focusing on coding and complex workflows, while Sonnet 4 serves as an all-purpose upgrade for developer tasks. Both models demonstrate improved performance on benchmarks like SWE-bench and MMMLU, with Opus 4 achieving 72.5% on SWE-bench and Sonnet 4 scoring 72.7%. The models feature expanded memory systems, parallel tool use, and "Thinking Summaries" for long reasoning chains.
Anthropic is activating its AI Safety Level 3 (ASL-3) standard for Claude Opus 4 to monitor inputs and outputs in real time, protecting against CBRN-related misuse and weight theft. The release also includes updated API features such as a code execution tool, MCP Connector for external systems, and prompt caching improvements.
Claude Code is now generally available for integration into IDEs like VS Code, allowing developers to implement feedback and fix errors directly within their workflows.