Anthropic Launches Claude Sonnet 4.5
Anthropic has officially launched Claude Sonnet 4.5, calling it the world’s best AI coding model.
The company says this new model brings a major leap in AI capability, with groundbreaking skills in coding, reasoning, and autonomous work.
The most striking feature: Claude Sonnet 4.5 can work on complex projects for more than 30 hours straight without losing focus — a dramatic jump from its predecessor, which lasted just seven hours.
What Makes Claude Sonnet 4.5 Stand Out
Anthropic says Sonnet 4.5 is now the top coding model worldwide, outperforming competitors on the SWE-bench Verified test, a benchmark for real-world software engineering.
In a live demo, Claude Sonnet 4.5 built a full chat app like Slack or Microsoft Teams, writing over 11,000 lines of code from start to finish without interruption.
This showcases a new level of independence and endurance for AI systems.
Smarter at Computer Use
One of the biggest upgrades is in computer use. On the OSWorld benchmark — which measures how well AI can understand and control software — Sonnet 4.5 scored 61.4%, up from 42.2% just months ago.
This means the model can “see” and navigate computer screens, click buttons, fill forms, and handle apps almost like a human user.
Anthropic designed Sonnet 4.5 as the strongest agent-building model yet, with improvements in memory, tool use, and multi-step reasoning.
It can manage several tools at once, delegate tasks to smaller agents, and coordinate complex workflows across coding, finance, cybersecurity, and research.
Real-World Uses
Coding Power: Companies report that Sonnet 4.5 can now handle long projects — from building apps to setting up databases and even running SOC 2 security audits.
Finance & Research: The model performs better in financial analysis, law, medicine, and STEM research, showing sharp reasoning and predictive analysis.
Cybersecurity: It can even deploy agents that patch software vulnerabilities automatically before hackers can exploit them. One firm said it reduced average vulnerability response times by 44%, with accuracy up 25%.
Safer and More Reliable
Anthropic also highlights safety. Sonnet 4.5 has fewer risks of “bad behaviors” like deception, power-seeking, or blindly agreeing with users.
It is more resistant to prompt injection attacks — where hackers trick AI into unsafe actions.
It’s released under AI Safety Level 3 (ASL-3), meaning strict filters protect against harmful content such as chemical or biological weapon discussions.
The company says false positives have dropped tenfold, making the model both safe and useful.
New Tools for Developers
Alongside the new model, Anthropic rolled out fresh tools:
Claude Code Enhancements: Features like checkpoints to roll back progress, an updated terminal, and a native VS Code extension.
Claude Agent SDK: Lets developers use the same infrastructure that powers Claude Code.
API Upgrades: New context editing, smarter memory, and cross-chat recall for long projects.
Direct Integration: The model can now create spreadsheets, slides, and documents inside chats. A Chrome extension is also available for premium users.
Availability and Pricing
Claude Sonnet 4.5 launched worldwide on September 29, 2025. Pricing stays the same as the older model — $3 per million input tokens and $15 per million output tokens.
Available on Claude.ai, iOS, Android, and APIs.
Integrated into Amazon Bedrock, Google Vertex AI, and GitHub Copilot.
Free users get limited access, while premium users get full use.
Developers can find it via API under the model ID: claude-sonnet-4-5.
Competition Heats Up
Claude Sonnet 4.5 enters a crowded field. It currently outperforms OpenAI’s GPT-5, GPT-5 Codex, and Google’s Gemini 2.5 Pro in key coding benchmarks.
Big names like Apple, Meta, GitHub, Canva, and Figma have already tested the model, praising its speed and quality.
Michael Truell, CEO of Cursor, called it “state-of-the-art coding performance,” while Jeff Wang of Windsurf said it represents “a new generation of coding models.”
News Gist
Anthropic launches Claude Sonnet 4.5, the world’s most advanced AI coding model, capable of working autonomously for 30+ hours on complex projects across coding, cybersecurity, and research.
FAQs
1.What is Claude Sonnet 4.5?
Claude Sonnet 4.5 is Anthropic’s latest AI coding model, designed to work autonomously for extended periods and handle complex coding and reasoning tasks.
2.What makes Claude Sonnet 4.5 different from previous models?
It can operate independently for over 30 hours, outperforming its predecessor Claude Opus 4.1, which lasted only seven hours, and generates high-quality code with advanced reasoning.
3.Which industries can use Claude Sonnet 4.5?
The model is suitable for software development, finance, cybersecurity, research, law, medicine, and other sectors requiring complex problem-solving and automation.
4.What new features come with Claude Sonnet 4.5?
Features include Claude Code enhancements, Claude Agent SDK, improved APIs with cross-chat memory, direct integration for documents and spreadsheets, and a Chrome extension for Max users.
5.Where and how can I access Claude Sonnet 4.5?
Available globally on Claude.ai, iOS, Android, Amazon Bedrock, Google Vertex AI, and GitHub Copilot. Free users have limited access; premium users enjoy full access.
6.Is Claude Sonnet 4.5 safe to use?
Yes, it comes with AI Safety Level 3 protections, filters for harmful content, improved resistance to prompt injection attacks, and safety controls for enterprise use.
