Patronus AI is Simulating the Entire World’s Intelligence and Unveiling its First Digital World Model for AI Agent Training and Simulation

Patronus AI, developing simulation research and infrastructure to accelerate progress toward human-aligned AGI, led by Anand Kannappan and Rebecca Qian, has announced its $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. The company is also releasing a preview of the first Digital World Model.

Background.

Since its founding, Patronus AI has developed AI evaluation research and products, including FinanceBench, Lynx, and Percival, with its platform adopted by leading enterprises and hundreds of thousands of developers. Since its Series A round, the company has focused on advancing frontier AI evaluation by moving beyond static benchmarks toward dynamic simulation environments, where AI agents can learn from complex, long-horizon tasks. Patronus AI believes these simulation-based environments will play a key role in the next generation of AI training and intelligence.

Simulating the Digital World.

Patronus AI is developing simulation environments that enable AI agents to learn complex, long-horizon tasks across digital workflows. Rather than relying solely on static datasets, these environments allow agents to practice and adapt within realistic scenarios, helping build the capabilities needed to operate across diverse software applications and enterprise use cases. The company views simulation infrastructure as a foundational technology for the next generation of AI systems.

Digital World Models.

Patronus AI is developing Digital World Models, language-based simulation models designed to predict environment behavior and guide AI agents across digital workflows such as coding, research, and communication. The technology is intended to support increasingly complex, long-horizon tasks while enabling scalable, self-improving simulation environments for training next-generation AI agents.

Benchmark Performance.

Patronus-DWM demonstrated leading performance across coding, dialogue, research, and general tool-use benchmarks under high-reasoning settings. The evaluations covered a broad range of public benchmarks spanning software engineering, AI research, graphical user interfaces, and enterprise tool use, highlighting the model's capabilities across complex digital workflows.