Anthropic has released the latest additions to its Claude model family "Claude Opus 4 and Claude Sonnet 4" marking practical improvements in AI-assisted coding, extended reasoning, and tool-enhanced workflows. Rather than positioning these models as breakthroughs, the release focuses on enhanced reliability, usability, and integration for real-world applications.

Key Highlights

Claude Opus 4: Opus 4 achieves strong results in software engineering tasks, scoring 72.5% on SWE-bench and 43.2% on Terminal-bench, which measure model effectiveness on real-world coding problems. It maintains stable performance on complex and extended tasks, with users like Cursor, Replit, and Rakuten noting improved multi-file code editing, sustained task focus, and internal validation in long-duration workflows.

Claude Sonnet 4: Sonnet 4 builds on the prior Sonnet 3.7 version with a measured performance gain, including a 72.7% SWE-bench score. It is being adopted for a range of development scenarios, offering a balance of capability and responsiveness. GitHub, Sourcegraph, and others report notable improvements in instruction following, code quality, and reduced navigation errors in multi-feature development.

Extended Tool Use (Beta): Both models now support integrated tool use, such as web search, during extended tasks. This allows for alternating between reasoning and tool-assisted thinking, potentially improving response quality in multi-step scenarios.

Parallel Execution and Memory: The models can now use tools in parallel and maintain working memory when given local file access. This enables them to store relevant information and build continuity across sessions—useful for complex workflows like refactoring or game strategy generation.

Claude Code General Availability: Claude Code, previously in preview, is now available to all developers. It integrates directly with VS Code and JetBrains IDEs, showing inline suggestions during editing. The accompanying Claude Code SDK allows teams to embed Claude into custom agent workflows. GitHub integration is in beta, offering support for pull request feedback, CI error handling, and review automation.

New API Features

Anthropic is releasing four new tools for developers via the Claude API:

Code Execution Tool

1. MCP Connector
2. Files API
3. Prompt Caching (up to 1 hour)

These tools are designed to support agentic workflows and long-context applications more efficiently.

Performance and Availability

Claude 4 models are accessible via the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI.

Pricing remains the same as previous versions:

1. Opus 4: $15 (input) / $75 (output) per million tokens
2. Sonnet 4: $3 (input) / $15 (output) per million tokens
3. Sonnet 4 is also available to free-tier users.

Evaluation and Responsible Deployment

Anthropic reports that Claude 4 models are 65% less likely to exploit task shortcuts or loopholes compared to Sonnet 3.7 in agent-based scenarios. Additionally, Developer Mode is available for teams needing deeper visibility into Claude’s thought process, offering full access to intermediate steps.

Rather than chasing headlines, Claude 4 delivers tangible improvements where it matters most: coding assistance, reasoning continuity, and development environment integration. These updates represent a thoughtful evolution of the Claude platform, with careful attention to safety, performance, and developer usability.