Anthropic makes Computer use, the browser tool, Skills API and Files API generally available

What’s changing

Anthropic announced on 20 August 2026 that four Claude Platform tools are now generally available (GA): Computer use, the Skills API, and the Files API, alongside the launch of browser use as a new tool in the computer use family, built specifically for agents working inside web applications.

The announcement opens by saying, plainly, that together these let you build agents that operate software, apply your team’s expertise, and return finished files.

What each of the four does

• Files API: stores documents for an agent to use. Upload once, then reference the file by file_id in later requests instead of re-sending it, and download files the agent itself creates, such as output from a skill or the code execution tool.
• Skills API: stores a workflow as a versioned folder of instructions, scripts, and templates. Claude loads it only when a task actually calls for it, instead of stuffing a long instruction set into every request.
• Computer use: lets Claude see the screen through screenshots, then click, type, and scroll like a person would. Built for software with no API to call.
• Browser use: builds on computer use specifically for web applications. The key change is how it targets things: instead of aiming at pixel positions on screen, it reads the structure of the page and acts on the specific element directly, which holds up much better when a layout changes.

The new computer use handles several actions per turn

The change inside computer use is that it can now take several actions in a single model call, instead of clicking once and looping back to ask the model again each time. Anthropic describes this as finishing tasks in fewer calls and less time.

A real case from Asteroid

Anthropic cites Asteroid, a company building agents for healthcare and insurance systems that have no API, testing the new computer use tool with these results:

• Their longest claims workflow went from 32 minutes down to 13 minutes.
• Cost per task dropped roughly 30% across the workflows they tested.
• Completion rate reached 100%.

Who this affects

This doesn’t directly touch anyone using Claude.ai or Claude Code day to day, since these are Claude API tools for people building their own agent systems. What changes is that all four have left beta: no beta header is required to call them anymore (a few SDKs, PHP among them, still reference a beta namespace in their sample code), and they’re ready to use as production components.

The writer’s take

I don’t think this touches regular Claude.ai or Claude Code users directly, but if you’re building your own agent system on the API, these are four pieces that are now solid enough for production, without worrying they’re still beta. The Asteroid numbers help too, they’re not just lab figures, they come from someone actually running this in production and seeing the result.

---

Details in this article come from Anthropic’s official announcement at claude.com/blog, published 20 August 2026, together with Claude Platform’s API documentation. Read the original post via the source link below.