Home / Directory / Data & labeling / Synthetic data / Patronus AI
Patronus AI
Training tasks and simulated software environments that large AI labs use to train agents with reinforcement learning, along with research on world models that predict how software responds to an agent's actions. Patronus AI started in 2023 with tools for testing large language model outputs, including Lynx, an open-source model that flags hallucinated answers, and Percival, which reviews an agent's step-by-step work for errors. Those evaluation products no longer appear on the company's website.
AI labs that buy reinforcement learning tasks, graders and simulated software environments to train agents on long, multistep work.
Teams that want a self-serve tool to test their own AI app, or buyers who need published prices before a sales call.
Verdict
Patronus AI fits AI labs and platform teams that want a vendor to build training tasks and simulated environments for agents that work through long, multistep jobs in software. Its simulations are most useful when an agent has to keep track of information that changes across several connected apps, which single-turn checks do not test. Teams that want a self-serve tool for scoring their own app's outputs will not find one on the Patronus site today, because the company no longer lists its evaluation products.
Score Breakdown
How Patronus AI scores in the categories that matter to its buyers.
Pricing
Patronus AI does not publish prices. Its pricing page, checked on October 23, 2026, asks AI labs to contact the company about training needs, project scope and pricing, so each project is priced through a sales conversation. The free Developer plan and the per-call evaluator API rates that used to appear on that page were removed between mid-September and mid-October 2026.
The Field at a Glance
Where Patronus AI ranks among Synthetic data vendors we reviewed, by Overall Score and relative typical engagement cost.
Patronus AI scores 6.7 in Synthetic data, under Anonos (7.7), K2view (7.3), and Gretel (7.1), and ahead of Tonic.ai (6.4). Patronus publishes no prices and scopes each project with the lab, so its cost position is an estimate toward the higher end of this set, not a list price.
Compared with …
- Patronus AI vs Braintrust 7.8/6.7
- Patronus AI vs Vellum 7.0/6.7
Use-case matrix
| Use case | Fit | Notes |
|---|---|---|
| RL training tasks with graders | Strong | Tasks come with expert-checked reference solutions, rubrics and graders. |
| Simulated software and computer-use environments | Strong | Ready-made environments cover software development, computer use and knowledge work, and Patronus also builds custom ones. |
| World-model environment generation | Mixed | The Digital World Model is a preview playground and API. |
| Privacy-safe copies of company data | Poor | Patronus builds agent training data, not stand-ins for production tables. |
| Self-serve eval for your own app | Poor | The evaluator API and free plan came off the pricing page in 2026. |
Review Excerpts
Below are excerpts from public reviews. Paid reviews and pay-for-play sites such as Clutch were excluded.
“Patronus helped our AI team find signals and patterns of error in our datasets. Their LLM Judges enabled us to triage errors and optimize our AI outputs in production settings.”
“Agent evals that aren't just string match finally showed where our tool-calling loop dies after step eight.”
“We run longer simulations before letting agents touch production tickets. Slower to ship, fewer weekend pages.”
“Setup cost is real. If you wanted a tracing dashboard with three charts, you'll feel oversold.”
Company Facts
- Founded
- 2023 (TechCrunch)
- Headquarters
- San Francisco, California, United States (TechCrunch)
- Total funding raised
- $70M (TechCrunch)
- Latest funding
- $50M Series B led by Greenfield Partners, June 2026 (Patronus AI)
- Key investors
- Greenfield Partners, Lightspeed, Notable Capital, Datadog, Samsung, Factorial Capital, Gokul Rajaram (Patronus AI)
- Website
- patronus.ai
Key Leadership
Anand Kannappan
Co-founder and CEO
Rebecca Qian
Co-founder and CTO
Methodology
This page is an independent evaluation of Patronus AI for buyers comparing options in synthetic data. AI Industry Reviews accepts no sponsorships, advertising, or pay-for-placement fees. Patronus AI did not pay for this review.
What we scored
The headline number is an Overall Score on a 0-10 scale. Eight criteria fall under it in two groups.
- Buyer outcomes (for synthetic data)
- RL tasks and environments
- World models and synthetic generation
- Research and benchmark record
- Customer evidence
- Company & commercial
- Innovation & product leadership
- Project management & communication
- Pricing
- Contract fairness
Pricing measures whether the price looks fair for the value delivered, including packaging and renewal friction that show up in real buying cycles.
Score Composition
| Input | Weight | What it covers |
|---|---|---|
| Reviews | 40% | A proprietary read of what practitioners say about likes, complaints, and day-to-day use, including public review sites, forums, and private chat rooms. Paid reviews and pay-for-play sites such as Clutch are out of scope. |
| Product | 35% | Hands-on look at screens and workflows. |
| Pricing | 15% | Whether the price looks fair for what you get. |
| Docs & training | 10% | Docs, tutorials, and training material. |
How we balanced the evidence
The Overall Score is the simple average of the eight criteria. Recommendation language follows that score and the fit pattern described above.
Scope
Patronus AI is graded here as synthetic data. Criteria scores can move as more review volume and product checks are added.