Patronus AI Raises $50M to Build Simulation Environments for AI Agent Testing

Quick Facts

• Patronus AI raised $50M in a Series B led by Greenfield Partners, bringing total funding to $70M.
• The company launched Digital World Models, large-scale simulations that train and evaluate AI agents across complex workflows.
• Revenue grew more than 15x over the past year, with clients including leading frontier AI labs and hyperscalers.

Patronus AI announced a $50 million Series B on June 25, bringing its total capital raised to $70 million. Greenfield Partners led the round, with participation from Notable Capital, Lightspeed, Datadog, Samsung, Factorial Capital, and angel investor Gokul Rajaram.

Alongside the funding, the company unveiled Digital World Models, a new class of simulation environments built to train and test AI agents on realistic software, research, and enterprise workflows. Patronus released a preview of its first Digital World Model and said it plans to expand its research and engineering teams while investing in the compute needed to run these simulations at scale.

The Problem With Benchmarks

Patronus AI was founded by former Meta AI researchers Anand Kannappan and Rebecca Qian. The team also includes engineers and researchers from Amazon AGI and Google, with backgrounds in LLM evaluation, AI alignment, and embodied agents.

CEO Kannappan was direct about the limits of existing evaluation methods. “Benchmarks were never the destination,” he said. “Static evaluations tell you whether a model can answer a narrow question in a controlled setting. They do not tell you whether an agent can navigate ambiguity, recover from failure, or operate reliably across long, unpredictable workflows.”

The urgency behind that argument is backed by industry data. AI agents fail on roughly 63% of complex multi-step tasks. Enterprise deployments show a 37-point gap between lab benchmark scores and real-world performance.

How Digital World Models Work

Patronus builds replicas of websites and internal systems. Inside these environments, agents are tested using reinforcement learning, which rewards successful task completion and penalizes errors. The company compares its method to how Waymo built synthetic worlds to expose autonomous vehicles to rare hazards before putting them on public roads.

“Manual review does not scale once AI systems begin operating across millions of workflows and decisions,” Kannappan said. “That is why simulations matter. They create environments where AI systems can be tested, improved, and supervised before failures happen in production.”

Patronus currently serves simulation environments for software engineering and finance use cases. Kannappan said there are “a ton more areas that are very non-verifiable or very hard to verify” the company plans to address.

Market Position and Competition

Patronus says it works with the majority of the world’s leading frontier AI labs and hyperscalers. The company’s earlier products, including FinanceBench, Lynx, and Percival, have been used by enterprises and hundreds of thousands of developers since the company launched less than three years ago. It currently employs 34 people, according to PitchBook.

Patronus operates differently from human-in-the-loop evaluation firms like Mercor and Surge. Its fully automated approach tests agent behavior without human involvement, which the company argues enables consistent testing at scale. Kannappan said Patronus is primarily competing against internal evaluation teams that AI labs have built themselves.

Glenn Solomon, managing director at Notable Capital, said demand for Patronus’ simulated environments is “nearly insatiable.” Itay Inbar, partner at Greenfield Partners, called the company’s work “one of the most important infrastructure problems in AI.”

What Comes Next

Patronus will use the Series B capital to grow its research organization, expand its engineering headcount, and build out the compute infrastructure required to train and serve Digital World Models at scale. Kannappan said the goal is to create environments where “an agent can run for 10 hours or 10 days or 10 weeks.”

Read more: Patronus AI grabs $50M in funding to stress-test AI agents in simulated environments