Patronus AI has raised a $50 million Series B led by Greenfield Partners and unveiled its first Digital World Models — large-scale simulation environments designed to train, evaluate and stress-test AI agents before they ever touch real systems.
A $50M bet on simulation
The round drew participation from existing backers including Lightspeed Venture Partners, Notable Capital, Datadog, Samsung and Factorial Capital, alongside angel Gokul Rajaram and a cohort of AI leaders. Founded in 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian, the San Francisco company has become a leader in AI evaluation and reliability testing, with revenue growing more than 15x over the past year as enterprises rush to deploy increasingly autonomous agents.
Digital worlds for agents to fail in
Patronus argues the next phase of model training will be defined by simulation rather than static datasets. Its Digital World Models are language diffusion world models that replicate websites and internal systems, creating environments where agents can practice long-horizon tasks, fail safely and improve through reinforcement learning that rewards successful completion and penalizes errors.
The Waymo analogy
The company compares its approach to self-driving cars: just as Waymo builds predictive world models so vehicles can rehearse scenarios they have never met on a real road, Patronus is doing the same for the digital world. But the digital domain, it argues, is a far larger problem — agents span coding, research, communication and tool use, each with its own logic, edge cases and failure modes, which is what makes scalable simulation infrastructure so critical.
Where the money goes
Patronus says it will expand its research organization, accelerate go-to-market and invest in the compute and infrastructure needed to train and serve Digital World Models at scale. Its Patronus-DWM preview reports leading results across coding, dialogue, research and general tool-use benchmarks, building on earlier products like FinanceBench, Lynx and Percival.
The raise reflects surging investor appetite for the infrastructure layer beneath agentic AI, echoing rounds such as General Intuition's $320M for gameplay training data, Generalist AI's $400M embodied-AI round and Baseten's $1.5B Series F for AI inference.
Reporting based on coverage from Patronus AI, TechCrunch and SiliconANGLE.