Anthropic Launches Claude Sonnet 4.5 with Best-in-Class Coding Abilities

Anthropic has released Claude Sonnet 4.5, positioning it as the world’s most capable coding model with significant improvements in software development, computer use, and complex reasoning tasks. The model demonstrates state-of-the-art performance on SWE-bench Verified evaluation and can maintain focus for over 30 hours on multi-step programming tasks.

Context and Background

Claude Sonnet 4.5 achieves 77.2% on SWE-bench Verified, substantially outperforming previous models in real-world software coding abilities. The model shows a 19.2 percentage point improvement over its predecessor on OSWorld computer use benchmarks, reaching 61.4% accuracy on tasks involving browser navigation and system interaction.

Early enterprise customers report dramatic productivity gains across diverse sectors. Cursor reports “state-of-the-art coding performance”, whilst GitHub highlights “significant improvements in multi-step reasoning and code comprehension” for Copilot’s agentic experiences. Legal technology firm Harvey notes the model excels at “analyzing full briefing cycles” and complex litigation analysis.

The release coincides with major product updates including checkpoints in Claude Code, a native VS Code extension, and the introduction of the Claude Agent SDK—the same infrastructure powering Anthropic’s own tools, now available to developers.

Looking Forward

Anthropic emphasises that Claude Sonnet 4.5 represents its “most aligned frontier model yet”, showing reduced concerning behaviours including sycophancy and deception whilst improving prompt injection resistance for agentic applications. The model operates under AI Safety Level 3 protections with enhanced security classifiers.

The company has also launched “Imagine with Claude”, a research preview allowing real-time software generation, demonstrating the model’s creative coding capabilities. Developer access remains at the same pricing as Claude Sonnet 4 (£2.30/£11.50 per million tokens), making advanced AI coding assistance more accessible to UK development teams.

Source Attribution:

Share this article