Key Takeaways
- Prentis is in talks to raise $100M at a $1B valuation, just months after its April launch
- The startup claims $50M in signed contracts and projects $75M ARR by Q3 2026
- Prentis says its Hive-32B model beats GPT-5.4 and Claude Opus 4.6 on computer-use benchmarks at 10x lower cost
Prentis, a new AI research lab training models to automate office workflows, is in talks to raise $100 million at a $1 billion valuation. The startup, co-founded by LinkedIn creator Reid Hoffman, Zynga founder Marc Pincus, and serial entrepreneur Ritankar Das, launched in April and already claims $50 million in signed customer contracts.
The round signals continued investor appetite for AI agents that can control computers, a category that Anthropic, OpenAI, and Mira Murati's Thinking Machines are all chasing. Prentis is betting it can win by running a smaller, cheaper model than frontier APIs.
What does Prentis actually build?
Prentis trains models to learn how knowledge workers navigate routine tasks, things like processing insurance claims, filing customs duty refunds, and hunting down paperwork across multiple systems. The goal is AI agents that can take over the mouse and keyboard.
According to investor materials obtained by TechCrunch, the company has signed contracts with a healthcare management organization, a manufacturer, and goods and clothing producers. Those contracts are worth up to $50 million total, with Prentis projecting an estimated $75 million annualized run rate by Q3 2026.
A caveat buried in the pitch deck: those figures reflect "estimated annualized value based on a contracted fee equal to 20% of savings realized," not recognized revenue. They're "performance-dependent and subject to final execution." In other words, Prentis gets paid a cut of whatever it saves its customers. If the agents underperform, revenue drops.
How does Prentis compare to OpenAI and Anthropic?
Prentis claims its Hive-32B model outperforms GPT-5.4 and Claude Opus 4.6 on two computer-use benchmarks: WindowsAgentArena, which measures end-to-end task completion on Windows apps, and ScreenSpot-v2, which tests a model's ability to find the right on-screen button.
The company's pitch is economics. It claims roughly 10x lower cost per task than frontier APIs, making it cheaper to deploy across everyday office workflows. TechCrunch hasn't independently verified the benchmark results.
| Model | WindowsAgentArena | ScreenSpot-v2 | Cost per task (claimed) |
|---|---|---|---|
| Prentis Hive-32B | Highest (per Prentis) | Highest (per Prentis) | ~10x lower than frontier APIs |
| OpenAI GPT-5.4 | Lower (per Prentis) | Lower (per Prentis) | Standard frontier API pricing |
| Anthropic Claude Opus 4.6 | Lower (per Prentis) | Lower (per Prentis) | Standard frontier API pricing |
Anthropic isn't just building its own computer-use models. It acquired Seattle startup Vercept earlier this year, folding in the founders and shutting down the product. That acquisition pattern suggests the big labs see this category as strategically important.
For founders building AI agents, Anthropic's official patterns offer a useful reference for agent architecture
Who are the founders behind Prentis?
Ritankar Das, the CEO, runs Titan, a holding company that builds and operates AI startups. Das graduated UC Berkeley at 18 as the school's youngest University Medalist in over a century, then earned a master's at Oxford before dropping out of an AI PhD at Cambridge to start Titan in 2014.
Titan's portfolio includes Tala Health, an AI-powered virtual care provider that raised a $100 million seed round last year, and Forta Health, an autism care startup that raised $55 million from Insight Partners in 2024. Disease prediction company Dascena, also from Titan, was acquired by CirrusDx in 2022.
For Hoffman and Pincus, Prentis is a side bet. Hoffman said last month he's stepping down from Microsoft's board after nearly a decade to go "founder mode" on Manas AI, an 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 now runs Reinvent Capital with Hoffman as a senior adviser and published a memoir last month. Neither founder appears to be working on Prentis full-time.
Can Prentis compete in a crowded market?
Prentis is betting that automating routine office tasks will surpass coding as AI's biggest use case. The thesis is plausible. Every company has armies of people filling out forms, reconciling data between systems, and chasing down documents. If agents can do that reliably, the TAM is enormous.
The challenge is execution. OpenAI, Anthropic, and well-funded startups like Thinking Machines are all building here. Prentis has hired over 25 employees, including researchers from OpenAI, DeepMind, Meta, Tencent, and Alibaba, according to its website. That's a solid team, but it's small compared to the frontier labs.
This week's AI funding shows continued investor appetite for both digital and physical AI agents
The performance-based revenue model is also a double-edged sword. It aligns Prentis with customer outcomes, but it means revenue depends on agents actually delivering measurable savings. If the technology underperforms in real-world deployments, the $75M ARR projection evaporates.
Logicity's Take
Prentis has the founder pedigree and the right thesis. Routine office work is tedious, expensive, and ripe for automation. But the $1B valuation on a three-month-old company with performance-dependent revenue is aggressive, especially when Anthropic and OpenAI are building competing products with billion-dollar R&D budgets. For founders in adjacent spaces, watch whether Prentis can convert those signed contracts into real, recognized revenue by year-end. That will tell us if smaller, cheaper computer-use models can actually win against frontier labs or if this category consolidates around the big players.
What happens next?
The $100 million round hasn't closed yet. Prentis declined to comment on the discussions. If the raise goes through, it would give the company enough runway to scale its team and compete for enterprise contracts against labs with far more resources.
The bigger question is whether "computer use" becomes a commodity feature built into every foundation model or a specialized capability that standalone companies can own. Prentis is betting on the latter. The next two quarters will show whether that bet pays off.
Frequently Asked Questions
What is Prentis AI?
Prentis is a new AI research lab training models to automate office workflows by controlling computers. It was co-founded by Reid Hoffman, Marc Pincus, and Ritankar Das in April 2026.
How much is Prentis raising?
Prentis is in talks to raise $100 million at a $1 billion valuation, according to sources familiar with the discussions.
What is Prentis's Hive-32B model?
Hive-32B is Prentis's computer-use model. The company claims it outperforms GPT-5.4 and Claude Opus 4.6 on benchmarks while costing roughly 10x less per task.
Who are Prentis's competitors?
Anthropic, OpenAI, and Mira Murati's Thinking Machines are all building AI agents for computer use. Anthropic acquired startup Vercept earlier this year in this category.
What customers does Prentis have?
Prentis has signed contracts with a healthcare management organization, a manufacturer, and goods and clothing producers, worth up to $50 million total.
Need Help Implementing This?
Building AI agents or exploring automation for your startup? Reach out to our team at Logicity.in for implementation guidance and vendor recommendations tailored to your stack and budget.
Source: Startups | TechCrunch / Marina Temkin
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
