Claude 4’s new models push the frontier of AI development with state-of-the-art coding, extended reasoning, and powerful agent capabilities—earning strong early praise from developers and researchers.
On May 22, 2025, Anthropic introduced the Claude 4 family of models—Claude Opus 4 and Claude Sonnet 4—marking a major leap in AI capabilities across coding, reasoning, and autonomous agent workflows.
Claude Opus 4 is positioned as the world’s best coding model, excelling in long-running, complex tasks. It leads industry benchmarks like SWE-bench (72.5%) and Terminal-bench (43.2%), with the ability to maintain performance over extended periods—up to several hours. Opus 4 features robust memory handling, especially when integrated with developer tools, enabling it to retain and reference information over time for improved coherence in tasks like code refactoring and gameplay strategy.
Claude Sonnet 4, while more lightweight, significantly outperforms its predecessor (Sonnet 3.7) in code generation and task following, earning a SWE-bench score of 72.7%. It has been praised for its ability to follow nuanced instructions, improve code quality, and reduce navigation errors in complex projects.
A key innovation in both models is “extended thinking with tool use”—a feature in beta that allows Claude to alternate between reasoning and tools like web search during a task. Both models can execute tools in parallel, access local files for memory, and follow instructions with greater precision.
Anthropic also rolled out Claude Code, now generally available, which integrates with development environments like VS Code and JetBrains. It supports native GitHub Actions and offers an SDK for custom agent development.
Initial reactions from developers and AI enthusiasts have been highly positive. Companies like Cursor, GitHub, Replit, and Rakuten have all endorsed Claude 4’s improvements in real-world software engineering. Online, the release has generated significant buzz, with many comparing Opus 4 favorably to competitors like GPT-4.1 and Gemini 2.5, particularly in agentic applications and developer productivity.
With pricing unchanged from the previous generation, Claude 4 models are now accessible via the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI.
