Anthropic Acquires Vercept to Push Claude Deeper Into Computer Use

What Anthropic bought

On February 25, 2026, Anthropic said it acquired Vercept to improve Claude's computer use capabilities. Anthropic describes computer use as the ability for Claude to work inside live applications the way a person at a keyboard would, taking multi-step actions across interfaces rather than stopping at text output or API-only integrations.

Anthropic said Vercept was built around the challenge of solving perception and interaction inside the same software humans use every day. The Vercept team, including co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick, will join Anthropic, and the company said Vercept will wind down its external product in the coming weeks as the team moves into Anthropic's roadmap.

The acquisition was linked directly to Anthropic's own model progress. The company said Claude Sonnet 4.6 raised performance on OSWorld, a widely used computer-use evaluation, from under 15% in late 2024 to 72.5% at the time of the announcement. Anthropic said the model is now approaching human-level performance on tasks such as navigating complex spreadsheets and completing web forms across browser tabs.

Why it matters

Computer use is becoming one of the most contested parts of the agent market because it changes models from analysts into operators. Once an AI system can interpret screenshots, move through software, and complete actions across tabs and tools, it becomes relevant for enterprise back offices, support operations, repetitive knowledge work, and older systems that do not expose clean APIs.

Anthropic's acquisition suggests that frontier labs increasingly view the interface layer itself as a core capability, not an optional addon. The company also pointed to Vercept as the latest in a sequence of targeted capability acquisitions. If Anthropic can combine stronger perception, safer action, and better reasoning, Claude becomes more relevant for work that requires actually doing tasks inside software, not only drafting instructions about what a human should do next.