Prentis, a computer-use AI lab co-founded by serial entrepreneur Ritankar Das with Reid Hoffman and Mark Pincus, is in talks to raise $100 million at a $1 billion valuation just three months after launching in April. The startup has already signed customer contracts worth up to $50 million and is projecting a $75 million annualized run rate by the third quarter of this year, according to investor materials. Those figures reflect a contracted fee equal to 20% of customer savings, not recognized revenue.
The pitch is narrower than the usual foundation-model story. Prentis is training models on how office workers actually move through documents, forms, and business software, then deploying agents that drive those same applications to handle insurance claims, customs duty refund exceptions, and other back-office paperwork. Early customers include a healthcare management services organization and manufacturers of goods and clothing.
The company's technical claim is that a small model beats big ones on the specific benchmarks that matter for this work. Prentis says its Hive-32B model outperforms OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on WindowsAgentArena, which scores end-to-end task completion on real Windows applications, and on ScreenSpot-v2, which measures whether a model can find the right on-screen control. The benchmark results have not been independently verified.
Key facts
- 01Prentis is in talks to raise $100M at a $1B valuation, three months after launching in April 2026.
- 02The startup has signed up to $50M in customer contracts and projects a $75M annualized run rate by Q3.
- 03Its Hive-32B model claims wins over OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on WindowsAgentArena and ScreenSpot-v2.
- 04Prentis says its model costs roughly 10 times less per task than frontier APIs.
- 05The lab has hired 25+ staff, including researchers from OpenAI, Google DeepMind, Meta, Tencent, and Alibaba.
The economic argument follows from the size. Prentis claims roughly 10 times lower cost per task than frontier APIs, which it says makes agent deployment viable across the kind of high-volume, low-margin workflows that don't pencil out at GPT-5.4 pricing. The company frames its projected run-rate numbers in its own deck as "performance-dependent and subject to final execution" — a hedge tied to the savings-based fee model.
“performance-dependent and subject to final execution”— Prentis, pitch deck disclosure
Das, 31, is also the founder of Titan, an operating holding company he started in 2014 after dropping out of a Cambridge AI PhD program where he had been a Gates Cambridge Scholar. He was UC Berkeley's youngest University Medalist in more than a century, graduating at 18 with a double major in bioengineering and chemical biology before earning a master's at Oxford. Das has described Titan as a deliberate throwback to a Berkshire Hathaway–style holding structure funded by its own exits rather than outside limited partners.
Titan's other bets sit mostly in healthcare AI. Virtual care provider Tala Health raised a $100 million seed round last year. Autism care startup Forta Health raised $55 million led by Insight Partners in 2024. Disease-prediction company Dascena was acquired by CirrusDx in 2022. Prentis is the group's first push into general-purpose enterprise automation.
For Hoffman and Pincus, Prentis is a side project layered on top of active operating roles elsewhere. Hoffman, the LinkedIn co-founder and Greylock partner, stepped down last month from Microsoft's board after nearly a decade to focus on Manas AI, his AI drug-discovery startup. He was an early OpenAI investor and co-founded Inflection AI with Mustafa Suleyman before Microsoft absorbed most of that team in 2024. Pincus, the Zynga founder, now runs Reinvent Capital with Hoffman as senior adviser and published a memoir last month.
The category is filling up fast. Anthropic, OpenAI, and Mira Murati's Thinking Machines Lab are all working on computer-use agents, and Anthropic has been acquiring talent directly — it bought Seattle computer-use startup Vercept earlier this year, folded in its founders, and shut down the product. Prentis has hired more than 25 employees so far, including researchers from OpenAI, Google DeepMind, Meta, Tencent, and Alibaba.
The risks are the ones every agent startup faces at this stage. Benchmark wins on WindowsAgentArena and ScreenSpot-v2 don't automatically translate into reliable production behavior across a customer's messy internal software. The 20%-of-savings pricing model puts revenue recognition at the mercy of whether customers agree the savings actually materialized. And the $50 million in signed contracts is a ceiling, not a floor — how much converts to cash depends on execution over the next several quarters.
The interesting bet inside Prentis is the small-model economics. If Hive-32B genuinely delivers frontier-competitive computer-use performance at a tenth of the cost, that inverts the pricing logic of the incumbents, whose per-token economics assume premium workloads. A $1 billion valuation on a three-month-old company is priced for that inversion to hold. The next twelve months will show whether the benchmark lead survives contact with production workflows — and whether founder-mode enthusiasm from Hoffman and Pincus translates into the enterprise sales motion this thesis actually requires.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
