Claude Sonnet 4.5 & Claude Code 2.0
Anthropic just launched its best AI model for development, along with Claude Code 2.0 that lets you undo changes instantly
Claude Sonnet 4.5 isn’t just another more powerful model
The numbers are impressive: 77.2% on SWE-bench Verified, the benchmark that measures real-world software coding abilities Introducing Claude Sonnet 4.5 \ Anthropic. But here’s what really matters for your daily work.
Claude Sonnet 4.5 can run autonomously for 30 hours Anthropic launches Claude Sonnet 4.5, its latest AI model that’s ‘more of a colleague’. That’s not a typo. Thirty complete hours maintaining focus on complex, multi-step tasks.
Remember when Claude Opus 4 came out four months ago? That model could work autonomously for 7 hours. This is 4 times longer.
An Anthropic researcher watched Claude Sonnet 4.5 not only build an application, but also stand up database services, purchase domain names, and perform a SOC 2 audit to ensure the product was secure Anthropic launches Claude Sonnet 4.5, its best AI model for coding | TechCrunch. All without human intervention.
But here’s the real secret Anthropic doesn’t shout in their press releases: the magic isn’t just in the model. It’s in the complete ecosystem they built around it.
Claude Code 2.0: The safety net you didn’t know you needed
Let’s talk about checkpoints, the most requested feature that finally arrived.
Every time you give Claude Code a prompt, it automatically saves your code state Enabling Claude Code to work more autonomously \ Anthropic. If something goes wrong — a refactoring that breaks tests, an “improvement” that actually makes things worse — you simply press Escape twice.
Boom. You’re back.
You can choose what to restore:
- Code only (keep the conversation)
- Conversation only (keep the changes)
- Both (complete reset)
Why is this revolutionary? Because you can finally explore without fear.
Before, giving an AI full access to your codebase was like handing your car keys to a novice driver. Now it’s like having a co-pilot with an instant “undo” button. The psychological difference is enormous.
The VS Code extension that changes everything
Claude Code now has a native VS Code extension in beta that brings capabilities directly into your IDE Enabling Claude Code to work more autonomously \ Anthropic. You no longer have to jump between your terminal and your editor.
You see Claude’s changes in real-time through a dedicated sidebar panel with inline diffs. It’s like doing code review, but with an AI that’s writing the code while you watch.
The terminal interface also got some love: improved status visibility, searchable prompt history with Ctrl+R. Small improvements that multiply when you use them 50 times a day.
And you know what else? It works with VS Code, Cursor, Windsurf, and VSCodium. You’re not locked into a specific tool.
Subagents, hooks, and background tasks: when one agent isn’t enough
This is where things get interesting for complex projects.
Subagents allow a main agent to delegate specialized tasks, like spinning up a backend API while the main agent builds the frontend Enabling Claude Code to work more autonomously \ Anthropic. Real parallel development.
Think of it as a team of developers working simultaneously on different parts of the project. Except it’s a single AI coordinating multiple versions of itself.
Hooks automatically trigger actions at specific points: run your test suite after code changes, lint before commits. Background tasks keep long-running processes like dev servers active without blocking Claude Code’s progress on other work.
Combined with checkpoints, you can confidently delegate broad tasks: extensive refactors, feature exploration, architecture migrations.
What does this mean for how you work?
Let’s be honest: every new AI model promises to revolutionize development. Most end up being an incremental improvement you use occasionally.
Sonnet 4.5 is different for one simple reason: the tools around the model are as important as the model itself.
Pricing remains the same as Claude Sonnet 4: $3 per million input tokens and $15 per million output tokens Introducing Claude Sonnet 4.5 \ Anthropic. You’re not paying more for significantly superior capabilities.
It’s already integrated into tools you probably use: GitHub Copilot, Cursor, Windsurf, Replit. Apple and Meta reportedly use Claude AI models internally Anthropic launches Claude Sonnet 4.5, its best AI model for coding | TechCrunch.
What makes this launch matter isn’t just the benchmarks. It’s that Anthropic understood something fundamental: developers don’t just need smarter models. They need confidence. They need control. They need to be able to experiment without destroying their work.
Claude Code 2.0’s checkpoints are that confidence materialized.
Your next step
Claude Sonnet 4.5 is available. If you’re a developer, you can access it via the Claude API with the model string claude-sonnet-4-5-20250929.
Download the VS Code extension from the marketplace. Update your Claude Code terminal installation. Try checkpoints on a project where you’d normally be afraid to make big changes.
And the next time you’re facing a massive refactoring that gives you chills, remember: you now have an undo button that works across 30 hours of autonomous work.
Relevant links:
- Official Claude Sonnet 4.5 announcement: https://www.anthropic.com/news/claude-sonnet-4-5
- Claude Code documentation: https://docs.claude.com/en/docs/claude-code
- Claude Agent SDK: https://www.anthropic.com/news/enabling-claude-code-to-work-more-autonomously
