Three-Month-Old AI Startup Prentis Is In Talks To Raise $100 Million At A $1 Billion Valuation

A company that launched in April 2026 is negotiating a deal that would value it at $1 billion. No product most people have heard of. No filed paperwork with regulators. Just a pitch deck and two people familiar with the talks.

Prentis is the company. It builds what it calls computer-use AI agents, software that watches a screen and operates it the way a human employee would, clicking through insurance claims, chasing down customs duty refund exceptions, filling forms nobody wants to fill. The company is in talks to raise $100 million at a $1 billion valuation, according to TechCrunch, which cited two people familiar with the discussions.

The names behind it are the reason this is getting attention. Prentis was co-founded by Ritankar Das, a 31-year-old serial entrepreneur who was UC Berkeley's youngest University Medalist in more than a century, alongside LinkedIn co-founder Reid Hoffman and Zynga founder Mark Pincus. Das also runs Titan, a holding company that builds and operates AI startups. He dropped out of an AI PhD program at Cambridge, where he'd been a Gates Cambridge Scholar, to found Titan in 2014.

The Numbers Behind the Valuation

Prentis says it has signed contracts worth up to $50 million with several customers, including a healthcare management services organization, a manufacturer, and clothing and goods manufacturers, according to TechCrunch. Investor materials obtained by TechCrunch project a $75 million annualized run rate by the third quarter of 2026.

That number sounds like revenue. It is not.

Prentis's own pitch deck states those figures reflect "estimated annualized value" based on a contracted fee equal to 20% of savings the company's software actually delivers for clients, not recognized revenue, and are "performance-dependent and subject to final execution." Prentis gets paid a cut of money it saves customers, and the $75 million figure assumes those savings materialize at scale. If the software underperforms, or clients dispute how much they actually saved, the number shrinks.

Unite.AI flagged an additional fact: because the round is still being negotiated, no Form D has been filed with U.S. regulators. That's the standard disclosure that pins down an actual investor list and a firm dollar amount once money changes hands. Right now, there is no such public record. The $1 billion figure exists in reporting based on people familiar with the talks, not in any filed document.

The Benchmark Claims Nobody Has Verified

Prentis says its in-house model, called Hive-32B, beats OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.6 on two computer-use benchmarks: WindowsAgentArena, which measures whether an AI can complete real tasks in Windows applications, and ScreenSpot-v2, which tests whether a model can find the right button or control on a screen.

The company also claims Hive-32B costs roughly 10 times less per task than frontier APIs from bigger labs, which it's pitching as the reason cash-conscious enterprise buyers should pick a smaller specialized model over a general-purpose one.

TechCrunch was explicit that it has not independently verified any of these benchmark results. They come from Prentis's own pitch deck. No outside lab or academic group is cited confirming the numbers.

A Crowded, Well-Funded Field

Prentis isn't alone in chasing this. Anthropic, OpenAI, and Mira Murati's Thinking Machines Lab are all building computer-use agents, according to one of TechCrunch's sources. Anthropic has already been buying talent directly, acquiring Seattle-based computer-use startup Vercept earlier this year and folding in its founders while shutting down the original product.

This matters because Prentis's entire pitch rests on being cheaper and more specialized than giants with vastly more capital and compute. AI Weekly's analysis put it plainly: if a smaller, cheaper model tuned for narrow office tasks can genuinely undercut hosted frontier APIs, the market could split into "narrow-and-cheap versus broad-and-expensive" camps. But if OpenAI or Anthropic decide to compress margins to compete on price, a standalone specialist has far less room to absorb the hit.

Prentis did not respond to TechCrunch's request for comment.

What's Actually Unresolved

The round hasn't closed. The $50 million in signed contracts hasn't been detailed in terms of how fast they convert to cash or how disputes over "realized savings" get settled. No independent party has run Hive-32B against GPT-5.4 or Claude Opus 4.6 to check Prentis's benchmark claims.

None of that means the deal won't happen, or that the technology doesn't work as advertised. The $1 billion number currently rests on projections the company itself labels as estimates, a set of investor conversations that haven't produced a public filing, and benchmark claims from a pitch deck that no outside evaluator has replicated. Whether any of that changes will show up the moment, if it happens, that Prentis files paperwork confirming money actually moved.