 |
| Claude Opus 4 logo displayed in Anthropic's announcement of their new AI model lineup. |
Anthropic announced the release of Claude 4 today, introducing two new AI models that the company says set new standards for coding, reasoning, and AI agent capabilities. The launch includes Claude Opus 4, which Anthropic claims is "the world's best coding model," and Claude Sonnet 4, positioned as a major upgrade to the previous generation.
Claude Opus 4 demonstrates strong performance on industry benchmarks, achieving 72.5% on SWE-bench and 43.2% on Terminal-bench. The model is designed for sustained performance on complex, long-running tasks that can continue for several hours with thousands of steps. Claude Sonnet 4 also shows competitive coding performance with a 72.7% score on SWE-bench, while maintaining efficiency for broader use cases. The model offers what Anthropic describes as an "optimal mix of capability and practicality."
 |
| Bar chart comparing Claude 4 models against other large language models on SWE-bench Verified, a software engineering benchmark. Claude Opus 4 and Sonnet 4 show leading performance scores. |
Several major technology companies have provided early feedback on the new models. GitHub plans to integrate Claude Sonnet 4 as the model powering the new coding agent in GitHub Copilot. Cursor called Opus 4 "state-of-the-art for coding" and highlighted improvements in complex codebase understanding. Replit reported enhanced precision and significant improvements for complex multi-file changes, while Block noted that Opus 4 was the first model to improve code quality during editing and debugging in their agent system. Rakuten validated the model's capabilities through a demanding seven-hour open-source refactor that ran independently with sustained performance.
Both Claude 4 models introduce several technical enhancements. Extended thinking with tool use allows the models to use tools like web search during their reasoning process, alternating between thinking and tool execution to improve responses. The models can also use multiple tools simultaneously through parallel tool execution, improving efficiency in complex workflows. When given access to local files, the models can create and maintain memory files to store key information, enabling better long-term task awareness and coherence. Additionally, the models are 65% less likely to use shortcuts or loopholes to complete tasks compared to the previous Sonnet 3.7 model.
 |
| Screenshot of Claude Opus 4's memory feature showing automatically generated notes about navigation and gameplay strategies while playing Pokémon Red, demonstrating the model's ability to maintain context over extended tasks. |
Alongside the model releases, Anthropic announced that Claude Code is now generally available after a research preview period. The platform now supports background tasks via GitHub Actions and includes new integrations with VS Code and JetBrains IDEs, displaying edits directly within files. New beta extensions allow Claude Code to integrate directly into development environments, with proposed edits appearing inline within familiar editor interfaces. The company also released an extensible Claude Code SDK for building custom agents and applications.
Claude Opus 4 is priced at $15 for input and $75 for output per million tokens, while Claude Sonnet 4 costs $3 for input and $15 for output per million tokens, maintaining consistency with previous model pricing. Both models are available through Claude Pro, Max, Team, and Enterprise plans, with Sonnet 4 also accessible to free users. The models can be accessed via the Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI.
The release represents Anthropic's continued push into the competitive AI development market, where coding capabilities have become a key differentiator among leading AI companies. The announcement comes as artificial intelligence companies increasingly focus on specialized capabilities like software development and autonomous agent workflows to distinguish their offerings in a rapidly evolving marketplace.