Anthropic has today introduced its latest Claude 4 models: Opus 4 and Sonnet 4.
Anthropic presents its new generation of AI models in an announcement. Opus 4 is the most powerful model to date and Sonnet 4 can better understand instructions than its predecessor Sonnet 3.7. The AI company itself says it has set new standards in advanced reasoning, coding, and agent behavior.
Long-term Operation
Claude Opus 4 is built to independently perform complex tasks over multiple hours. According to Anthropic, AI agents running on Opus 4 can make decisions, process information, and perform actions autonomously without human input. This progress comes from improved memory training, where the model creates its own “notes” to maintain context, even when it needs to clear its memory.
Sonnet 4, the successor to Sonnet 3.7, was trained to deviate less from instructions. The generation of additional or irrelevant code has reportedly been reduced by 80 percent, making Sonnet easier to control and manage, especially in programming applications. Both models are hybrid, meaning they can both reason in steps and provide direct answers.
New Tools for Developers
Finally, Anthropic makes Claude Code generally available. This developer tool runs in a terminal or code editor and can combine natural language with complex codebase interactions.
Four additional functions are now available via the Claude API: executing code, prompt caching, and support for MCP. Additionally, Claude can now alternate between reasoning and using tools such as web search.