Anthropic Launches Claude Sonnet 4.5: The AI Coding Model That Works for 30+ Hours Autonomously

Anthropic just launched Claude Sonnet 4.5, and this AI coding assistant is turning heads across Silicon Valley. The company is calling it "the best coding model in the world" and they've got some impressive benchmarks to back up that claim.

Claude Sonnet 4.5 Key Features and Performance

The new AI model achieved a 77.2% score on SWE-bench Verified, a benchmark that tests real-world software coding abilities. That's better than most human engineers and represents a significant leap in AI coding capabilities.

What Makes Claude Sonnet 4.5 Different:

• 30+ hours of autonomous work (up from 7 hours with Claude Opus 4)
• 77.2% SWE-bench Verified score (industry-leading performance)
• Advanced computer use with 61.4% on OSWorld benchmark
• Production-ready applications, not just prototypes

During trials, the AI model built entire applications autonomously, set up databases, purchased domain names, and conducted security audits, all without human intervention.

Advanced Computer Use Capabilities

Claude Sonnet 4.5 doesn't just write code, it can actually navigate websites, fill spreadsheets, and complete real-world computer tasks. The model scored 61.4% on OSWorld, compared to 42.2% four months ago with the previous version.

The new Claude for Chrome extension demonstrates these capabilities, showing the AI browsing sites, filling in data, and completing multi-step workflows autonomously.

AI Safety Improvements in Claude Sonnet 4.5

Anthropic emphasizes that Claude Sonnet 4.5 is their "most aligned model yet," with significant safety improvements:

• Reduced sycophancy (telling users what they want to hear instead of the truth)
• Lower deception rates compared to previous models
• Better resistance to prompt injection attacks
• Decreased power-seeking behaviors

These safety enhancements make Claude Sonnet 4.5 more reliable for enterprise applications where accuracy and security matter most.

New Developer Features and Tools

Anthropic released several powerful tools alongside Claude Sonnet 4.5:

Claude Code Checkpoints: Save progress and roll back to previous states during development

VS Code Extension: Native integration for popular code editors

Claude Agent SDK: The same infrastructure powering Claude Code, now available for developers to build custom AI agents

"Imagine with Claude" (Beta): Experimental feature that generates software on the fly with no pre-written code (currently limited to Max subscribers for five days)

Claude Sonnet 4.5 vs. Competitors (OpenAI GPT-5, Google Gemini)

The AI arms race continues heating up. Recent studies showed that Claude Sonnet 4.1 outperformed GPT-5, Gemini, and Grok on real-world job tasks. Anthropic claims 4.5 represents a significant leap beyond that.

Major platforms are already integrating the new model:

• GitHub Copilot is rolling out Claude Sonnet 4.5
• Cursor CEO called it "state-of-the-art coding performance"
• Windsurf described it as a "new generation of coding models"

Claude Sonnet 4.5 Pricing and Availability

Despite significant performance improvements, Claude Sonnet 4.5 maintains the same pricing as its predecessor:

• $3 per million input tokens
• $15 per million output tokens
• Additional savings available with prompt caching (up to 90%) and batch processing (50%)

Where to Access Claude Sonnet 4.5:

• Claude.ai (web, iOS, and Android) - available on free and paid tiers
• Claude API with model string 'claude-sonnet-4-5-20250929'
• Amazon Bedrock
• Google Cloud's Vertex AI
• GitHub Copilot (rolling out to Pro, Business, and Enterprise)