Anthropic just dropped Claude Sonnet 4.5, and it’s coming in hot: faster, sharper, and already topping the coding and agent leaderboards. The company is calling it its most capable and most aligned model yet—and it’s rolling out across all Claude products today, at the same price as Sonnet 4.
The Big Claim: Best Coding Model in the World
Claude 4.5 is now #1 on SWE-bench Verified (77.2%, hitting 82% with high-compute tricks) and leads OSWorld at 61.4%—a massive jump from Claude 4’s 42.2% just a few months ago. Anthropic says the model can stay locked in on 30+ hour multi-step coding tasks without losing the plot.
In other words: this is the new benchmark for long-haul, complex software work.
Major Upgrades Across Claude Code
Developers get a pile of long-requested features:
- Checkpoints: save + roll back your coding sessions instantly.
- A polished terminal interface and a native VS Code extension.
- Context editing + memory for massive agent workflows.
- Direct code execution + file creation inside Claude apps (docs, sheets, slides).
- The Claude for Chrome extension finally rolling out to Max waitlisters.
And for builders who want to roll their own? Anthropic just open-sourced its playbook with the Claude Agent SDK—the same infrastructure it uses internally to power Claude Code.
Safer, Sharper, Less Weird
Anthropic is doubling down on alignment. Sonnet 4.5 ships under ASL-3 protections, with tighter filters for high-risk content and big improvements in resisting prompt injection (the bane of agentic AIs). Misaligned behaviors—like sycophancy, deception, and power-seeking—have dropped significantly compared to past models.
A Fun Extra: Imagine with Claude
For the next five days, Max subscribers can try Imagine with Claude, a real-time software-generation demo where the model builds from scratch, adapting as you interact. Think of it as a sandbox to see just how far Sonnet 4.5 can stretch.
Same Price, More Power
- Available today via API (
claude-sonnet-4-5), Claude apps, and Claude Code. - Pricing stays flat at $3 / $15 per million tokens.
- All the dev-focused upgrades (including the Agent SDK) ship today.
Bottom line: Claude Sonnet 4.5 isn’t just an incremental update—it’s Anthropic planting a flag. Stronger coding, stronger agents, stronger alignment. And at this point, the message is clear: Claude isn’t chasing anymore—it’s leading.
