Anthropic acquires Vercept to improve Claude’s computer use
Anthropic is acquiring Vercept to improve Claude’s ability to interact with applications like a person. The deal follows Claude Sonnet’s improvement from less than 15% to 72.5% on the OSWorld test.
Anthropic has acquired Vercept to strengthen Claude’s ability to use computer applications as a person would at a keyboard. The deal brings Vercept’s team into Anthropic, while its standalone product will stop working over the coming weeks.
The so-called computer use capability allows Claude to do more than generate text or code. It can also interact with open applications and complete multi-step tasks. For example, it can move between browser tabs, work with spreadsheets, or follow a workflow that combines several tools.
This matters because some problems cannot be solved by writing code. An assistant may need to interpret what appears on screen, decide what action to take, and check whether the result is correct. Vercept had spent years working specifically on those two capabilities: perceiving what is happening in an interface and acting within it.
Claude improves at computer tasks
Anthropic links the acquisition to recent progress in Claude Sonnet 4.6. In OSWorld, an evaluation used to measure whether AI systems can complete tasks on a computer, Sonnet models went from achieving less than 15% success at the end of 2024 to reaching 72.5% at the time of the announcement.
According to Anthropic, Sonnet 4.6 is approaching human performance on tasks such as:
- Navigating complex spreadsheets.
- Completing web forms spread across multiple tabs.
- Running processes that require several actions within an application.
The figure does not mean Claude can use any program without errors. It comes from a specific evaluation and reflects an improvement in certain types of tasks. Even so, it shows that AI assistants are moving from answering questions toward carrying out work inside the tools you already use.
What changes for you
If this capability continues to improve, Claude will be able to handle more complete parts of digital work: gathering information, moving it into a spreadsheet, filling out a form, and checking the result, all within real applications.
That also makes human controls more important. A system that can act on your behalf needs clearly defined permissions, supervision, and a clear way to review its decisions before they have consequences.
The Vercept team, including co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick, will join Anthropic. The company presents the deal as part of its strategy to add teams with technical expertise and a safety-focused approach. It is Anthropic’s second recently announced acquisition, after Bun.
What to watch now is whether the technology moves from demonstrations and controlled tests to everyday tasks with fewer errors. Claude’s next advance will not only be understanding your instructions better, but completing more work inside the applications you already use.