Patronus AI

Patronus AI builds simulation infrastructure and Digital World Models that train AI agents on realistic tasks — research, software development, customer service, product navigation, and finance — through high-quality interactive simulations rather than static training data. Its research models include Lynx (hallucination detection), FinanceBench (financial Q&A evaluation), BLUR (tip-of-the-tongue retrieval moments), and GLIDER (reasoning chain evaluation), and the company documents 30-40% model performance improvements on long-horizon tasks from training against its simulations. A public Playground lets you test the Digital World Model directly. The company recently closed a $50 million Series B round — a real, checkable funding signal — though pricing isn't disclosed on the main page. Good fit for a team training or evaluating AI agents on long-horizon, multi-step tasks that wants realistic simulation environments rather than static benchmark datasets.