Anthropic · 2024 · 10 · 22 · Tool · ~1 min read
Claude 3.5 Sonnet (new) + Computer Use
What's actually new
- Computer Use. Claude can take a screenshot of your screen, decide what to click, type things, and try a task on its own. In a sandbox for safety.
- The new Sonnet got better. Same name, same price, better at coding and using tools.
- First major lab to ship an AI that controls a real on-screen interface, not just calls things behind the scenes.
If you want more
Worth knowing
- Demos are curated. In real use, Claude makes typing mistakes, gets confused by pop-ups, and sometimes clicks the wrong button.
- The naming is messy. Anthropic called this 'new Sonnet 3.5' without bumping the version number, which confused everyone tracking the API.
Who should care
Developers building 'AI that does things' apps. Anyone trying to figure out if 'AI does my work' is real or marketing. Teachers needing concrete examples of what AI can and can't do today.
What to do about it
Watch the demo. Then try a small task yourself before you believe the hype. Don't let Computer Use anywhere near your real passwords or production systems yet.
Honest take
Computer Use is the first version of something that will eventually feel as obvious as autocomplete. Right now it's a brittle preview — slow, error-prone, expensive — but the direction is clear. The quieter story is the upgraded Sonnet itself: it's the model most people use every day, and it quietly got better at coding and tool use, which matters a lot more than the splashy demo.
Sources
Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com