Claude computer use, browser tool, Skills and Files APIs reach GA
- Source
- x.com
- Date
Computer use, the browser tool, the Skills API, and the Files API are now generally available on the Claude Platform. Automate work in applications that have no API with fewer round trips per task, and build Claude Managed Agents on versioned skills and reusable files.
Context
Anthropic made four -building features generally available on the Claude API: computer use, the new browser use tool, Agent Skills with its Skills API, and the Files API. For the three that were previously in beta, general availability means developers no longer send the beta headers that switched each feature on, according to Anthropic's August 19 release notes. Requests that still use the old beta versions keep working, and for the Files API they keep the old response format, so existing integrations do not break. The browser use tool launched the same day: a toolset for a browser the developer's own application hosts, which reads page structure, fills forms and manages tabs.
The most concrete change is in computer use, where Claude operates a desktop by looking at screenshots and requesting mouse and keyboard actions that the developer's application carries out. With the new computer_toolset_20260801 version, Claude can return a short sequence such as click, type and screenshot in a single response; the application runs the actions in order, stops at the first failure, and sends all the results back together, as the computer use documentation explains. ClaudeDevs says in the follow-up post that this replaces one action per round trip, and that early-access customers needed 20 to 40 percent fewer round trips per task.
That figure counts exchanges between the model and the application, not elapsed time or spend. ClaudeDevs says fewer round trips make tasks faster and cheaper but gives no measurement of either. The batching also has a safety consequence the documentation spells out: because one turn can complete a multistep action, an application that asks a person to confirm consequential actions has to check each action in a batch before running it. Moving to the new version changes the request shape, and the documentation says that on the Claude API, Claude 5.5 and later models accept only the new toolset.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…







