Prentis is in discussions to raise $100 million at a $1 billion valuation, but the financing has not been announced as agreed or closed. The computer-use artificial-intelligence laboratory was launched in April by Ritankar Das, Reid Hoffman and Mark Pincus, so the proposed valuation is being tested only months after its formation.
The company trains models to learn how office employees navigate documents, websites and software systems, then builds agents intended to execute the same workflows. Its examples include processing insurance claims and handling exceptions in customs-duty refunds, tasks that require moving information across several interfaces rather than producing a single text response.
Prentis has signed customer contracts described as worth up to $50 million, involving a healthcare-management organization, a manufacturer and producers of goods and clothing. Investor materials project a $75 million annualized run rate by the third quarter, giving the fundraising case a commercial component in addition to the laboratory's model claims.
Those figures are not equivalent to booked revenue. Prentis' contracts use a fee equal to 20% of the savings actually realized by customers, and the projected annualized value is performance-dependent and subject to final execution. The gap between maximum contract value, annualized estimates and recognized revenue is therefore central to assessing the proposed valuation.
On the technical side, Prentis says its Hive-32B model outperforms OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on two computer-use benchmarks. WindowsAgentArena measures end-to-end task completion in real Windows applications, while ScreenSpot-v2 tests whether a model can identify the correct on-screen control. The results have not been independently verified.
The company also claims a cost per completed task roughly one-tenth that of frontier-model APIs. That assertion matters because savings-based pricing only works if the agent can complete workflows reliably at a cost well below the value it creates; a smaller model may improve margins and deployment speed, but benchmark leadership does not by itself establish performance in each customer's software environment.
Competition is already substantial: OpenAI, Anthropic and Thinking Machines Lab are developing agents that operate computers, while Anthropic acquired specialist startup Vercept and shut its product as it absorbed the founders. Prentis has assembled more than 25 employees, including alumni of OpenAI, Google DeepMind, Meta, Tencent and Alibaba, but it is entering a market where both capital and experienced teams are concentrated.
The next milestones are a signed financing, disclosure of its investors and evidence that customer deployments convert projected savings into collected fees. Until then, the $100 million round, the $1 billion valuation and the $75 million annualized run rate remain different stages of the same proposition—not completed financing, current market value and recognized revenue.
