Anthropic has introduced two new AI models: Claude Opus 4 and Claude Sonnet 4. The company contends that these are among the best out there, but that their chief value is in their logical prowess. Both models are fine-tuned for coding and agent-style reasoning. Claude Opus 4 is the best model yet from Anthropic, and it’s aimed at developers working on long, intricate projects.
According to Anthropic’s blog, “Claude Opus 4 is the world’s best coding model, delivering consistent performance on lengthy tasks and agent workflows.” Claude Sonnet 4, meanwhile, is a more useful, streamlined advance on Claude Sonnet 3.7, this time available as part of the free tier, with Opus 4 still only available to paid subscribers.
Table of Contents
Anthropic unveils Claude Opus 4 and Sonnet 4 AI Models with Improved Coding, Memory, and Tool-Use Features
A major success is the result of Claude Opus 4 being high on the coding benchmarks, with a mark of 72.5% in the SWE bench and 43.2% in the Terminal bench. This means the model can withstand hours of high performance to accommodate your tough projects. Claude Sonnet 4 also got better, scoring 72.7% on SWE-bench, balancing speed and accuracy well for common uses, but it’s less powerful than Opus 4.
Each is equipped with “extended thinking” and can access tools, such as the ability to pause reasoning and bring in tools like a web search or code execution, and then continue his or her thought process.
They can also help execute several tool-related functions at the same time in order to help manage multi-faceted workflows more efficiently. New memory capabilities enable the models to access local files, extract important information, and retain it for future use, improving long-term memory and task continuity.
Anthropic also rolled out four new API features: a code execution tool, a Multi-Component Programs (MCP) connector, a Files API, and prompt caching for up to an hour, helping developers create more capable AI agents.
Claude Opus 4 excels in memory during agent tasks, demonstrated by navigating a Pokémon game with file access, maintaining context, and reducing shortcut-taking tendencies seen in earlier models. To improve transparency, Anthropic added “thinking summaries”—brief explanations of the model’s reasoning—generated by a smaller AI, while full reasoning chains remain available in Developer Mode.
FAQs
What tasks is Claude Opus 4 best for?
It’s ideal for complex, long-running coding and agent workflows requiring sustained focus.
Can free users access both models?
Free users get Claude Sonnet 4, while Claude Opus 4 is for paid subscribers only.
