Claude Sonnet 4.5 Launched: What's New With Anthropic's Latest AI Model?

Anthropic has unveiled Claude Sonnet 4.5, its most advanced AI model to date. The release marks a significant leap in AI capabilities, emphasising practical applications in software development, business workflows, and specialised industries like finance and cybersecurity. Available immediately to all users at no additional cost over its predecessor, Sonnet 4.5 integrates seamlessly with enterprise tools, including Amazon Bedrock and GitHub Copilot, to streamline real-world productivity.

Claude Sonnet 4.5 model tops the SWE-bench Verified benchmark at 77.2 percent, the company claims, outperforming rivals in generating high-quality code, identifying improvements, and executing multi-step software projects. It is said to maintain focus autonomously for over 30 hours on intricate tasks—more than quadruple the endurance of its May predecessor, Claude Sonnet 4.

Sonnet 4.5 also excels in computer interaction, scoring 61.4 percent on the OSWorld benchmark for real-world tasks like browser navigation and spreadsheet manipulation. This enables it to handle everything from vulnerability patching in cybersecurity to predictive financial analysis, delivering consistent results across extended workflows. Early users in STEM, law, and medicine report "dramatically better" domain-specific reasoning, making it feel "more like a colleague" for problem-solving.

Claude Sonnet 4.5: New Features

New features amplify its agentic potential _ Checkpoints in Claude Code for instant rollbacks, a refreshed terminal interface, and a native VS Code extension. In the Claude apps, users can now execute code and generate files like spreadsheets or documents mid-conversation. For developers, the Claude Agent SDK provides the infrastructure Anthropic uses internally, tackling challenges like memory management and subagent coordination for custom agents beyond coding.

Claude Sonnet 4.5: Strategic Integrations with Amazon Bedrock and GitHub Copilot

Anthropic's push for enterprise adoption shines through in its platform integrations. In Amazon Bedrock, Claude Sonnet 4.5 debuts with tailored enhancements for building production-ready agents via AgentCore. Key API upgrades include smart context window management—preventing abrupt cutoffs by signaling limits mid-response—automatic tool-use cleanup to trim token costs, and cross-conversation memory via local files for persistent, personalized interactions.

On the developer front, Claude Sonnet 4.5 enters public preview in GitHub Copilot for Pro, Pro+, Business, and Enterprise users. Accessible via VS Code (in chat, edit, and agent modes), github.com, mobile apps, and the Copilot CLI, it boosts tool orchestration and context editing. Administrators can enable it organisation-wide, while individual users can bring their own API keys in VS Code. GitHub reports "major upgrades" in domain-specific tasks, making it ideal for collaborative coding.

Claude Sonnet 4.5: Pricing

Priced at $3/$15 per million input/output tokens via the Claude API, unchanged from Sonnet 4, Claude Sonnet 4.5 is recommended as a drop-in upgrade for all use cases.

Published on: Tuesday, September 30, 2025, 11:32 AM IST