# Patronus AI

> Patronus AI builds software and research tools for testing, debugging and training generative AI systems. Its evaluation platform scores model outputs, its Percival product analyzes agent traces, and its emerging Digital World Models simulate software environments for agent practice.

- **Founded:** 2023
- **Headquarters:** San Francisco, United States
- **Founders:** Anand Kannappan (Co-founder and CEO), Rebecca Qian (Co-founder and CTO)
- **Products:** FinanceBench, Patronus Evaluators and Experiments, Lynx, Multimodal LLM-as-a-Judge, Percival
- **Notable:** Raised $70 million in disclosed funding through its June 2026 Series B., Published FinanceBench, a financial QA benchmark whose full version contains 10,231 question-answer-evidence sets., Nova AI reported reducing review of an agent trace from about one hour to one minute with Percival.

## Products & services

- **FinanceBench** — Financial-document question-answering benchmark with 10,231 question-answer-evidence sets in its full version.
- **Patronus Evaluators and Experiments** — API, SDK and interface for scoring AI outputs, comparing models and testing prompts against datasets.
- **Lynx** — Specialized model for detecting hallucinations in retrieval-augmented answers.
- **Multimodal LLM-as-a-Judge** — Evaluation for image-to-text outputs, including caption relevance and accuracy.
- **Percival** — Agent debugger that analyzes traces for planning, tool-use and execution failures and suggests fixes.
- **Digital World Models** — Preview of generated digital environments for agent training and simulation.

## Achievements

- Raised $70 million in disclosed funding through its June 2026 Series B.
- Published FinanceBench, a financial QA benchmark whose full version contains 10,231 question-answer-evidence sets.
- Nova AI reported reducing review of an agent trace from about one hour to one minute with Percival.
- Gamma reported saving more than 1,000 hours of manual evaluation per month using Patronus Judges.
- Released a preview of a Digital World Model in June 2026.

## Latest updates

- **2026-06** — Announced a $50 million Series B led by Greenfield Partners and previewed a Digital World Model.
- **2026-02** — Published research on generative simulators for self-adapting intelligence.
- **2025-05** — Published case studies of Percival use at Nova AI, Weaviate and Emergence AI.
- **2025-03** — Launched multimodal LLM-as-a-Judge evaluation for image-to-text applications.

## Links

- Website: https://www.patronus.ai
- LinkedIn: https://www.linkedin.com/company/patronus-ai-inc
- Twitter/X: https://twitter.com/PatronusAI
- GitHub: https://github.com/patronus-ai

---

Profile page: https://yespress.io/patronus-ai
Published by YesPress — https://yespress.io
Last updated: 2026-09-29