Announcing our $50M Series B to Simulate the Entire World’s Intelligence and Unveiling our First Digital World Model for AI Agent Training and Simulation
Announcing our $50M Series B to Simulate the Entire World’s Intelligence and Unveiling our First Digital World Model for AI Agent Training and Simulation
Today, we’re excited to announce our $50M Series B led by Greenfield Partners, with participation from Lightspeed Venture Partners, Notable Capital, Datadog, Samsung, Gokul Rajaram, Factorial Capital, and a cohort of AI leaders across the labs and neolabs. We are also releasing a preview of the first Digital World Model.
How We Got Here
Since we started as a company, we’ve developed the earliest and most influential research and products in AI evaluation, like FinanceBench, Lynx, and Percival. Our products have been used by leading enterprises and hundreds of thousands of developers in the past few years.
And since our Series A, we've been working to define what it means to build at the frontier. We're living through an unprecedented shift, one that's moving away from static evaluation datasets toward dynamic environments.
The next phase of LLM training will be defined by simulations: carefully designed environments where agents can practice, fail, and learn from long-horizon tasks that never appear in static text corpora. These simulations are the most important infrastructure driving the next frontier of intelligence.
Simulating the Entire Digital World
Since childhood, every one of us has accumulated thousands of hours interacting with digital worlds: navigating software, writing documents, sending messages, managing workflows. We have a deep, intuitive understanding of how these environments work. Agents don't, and they need it to improve in long-horizon settings. If we want an agent that can operate like a human over the course of a year, it has to develop the same grounding humans have built since day zero. At scale, the only way to develop that grounding is through simulation.
Consider self-driving cars. Waymo can't physically drive every route in the world, so instead they build world models: predictive models that capture the latent dynamics of the physical world, from unexpected pedestrians to unusual weather. They simulate reality so the car can learn to handle scenarios it has never encountered on an actual road.
That’s exactly what we are doing as well, but for the digital world. And the digital world is, in many ways, a far bigger problem. Waymo solves one type of problem in the physical world: driving. Agents, by contrast, span countless domains and capabilities. Every domain has its own logic, its own edge cases, its own failure modes. That is what makes this problem so much larger, and what makes simulation infrastructure so critical.
Preview of our First Digital World Model
We are building Digital World Models – language diffusion world models that predict realistic environment behaviors and steer agent actions across digital workflows. We build world models for the digital world, for agents that navigate code, research, communication and more.
Digital World Models help us push the frontier of ultra long horizon workflows. By making it possible to target agent behaviors and scale environment building, these models unlock a new class of self-improving environments, building on the direction we first explored in our Generative Simulators work.
Patronus-DWM leads on Coding, Dialogue, Research, and General Tool Use, with all models run at high reasoning. Domains span public benchmarks across coding (InterCode, CoderForge, SWE-smith), dialogue (τ-bench), research (DeepResearchQA, OpenResearcher), GUI, and general tool use (OccuBench, API-Bank, BFCL-v4, Toolathlon, Pandora).
What’s Next
With the new funding, we plan to expand our research organization, accelerate GTM, and invest in the compute and infrastructure required to train and serve Digital World Models at scale.
We are hiring across a range of roles: Applied Research, Engineering, and more. If our mission resonates with you, we’d love to chat!
We always believe that the best is yet to come.
"Do not go gentle into that good night,
Rage, rage against the dying of the light"
- Dylan Thomas (1954)
Digital World Model Links
Playground: https://dwm.patronus.ai/playground
Documentation: https://dwm.patronus.ai/documentation
Research Paper:
