Catalog
Resources
Guides, comparisons, definitions, open datasets and research from Patronus AI, in one place.
01
Solutions
What we build and sell, each page at its own root URL.
02
Comparisons
How RL environment companies compare, with a source for every fact.
Best RL Environment Companies in 2026
Nine RL environment companies compared for 2026, from research labs and expert data firms to open hubs, with what each sells and who it suits.
Best Scale AI Alternatives for RL Environments in 2026
Nine Scale AI alternatives for RL environments in 2026, from expert data companies to research labs and open hubs, with what each sells and who it suits.
03
Guides
Long-form guides to RL environments, world models, agents and evaluation, each with chapters.
RL Environments: Tutorial & Examples
The fundamentals of reinforcement learning environments and how they let AI agents learn from trial and error.
AI World Models: What They Are and How They Work
How AI world models predict environment responses to actions, so agents can train without real systems.
AI Agent Development, Evaluation, and Optimization
How agents use language models, tools and memory to do tasks on their own, and how to evaluate them.
LLM Testing: The Latest Techniques & Best Practices
Model-centric and application-centric evaluation of large language models, with the tools and data for each.
AI Reliability: Tutorial & Best Practices
Correctness, consistency and availability, and how to assess and improve LLM and RAG systems on each.
04
Glossary
Plain definitions of RL environment and post-training terms, each tied to our research.
- 60 terms
RL environment and post-training glossary
Definitions from verifiers and pass@k to reward hacking, in five themes and A to Z.
What is a verifier in RL?
A check, usually code, that decides whether an agent’s answer or final state is correct.
Pass@k and RL difficulty calibration
The chance that at least one of k attempts succeeds, and how to use it to calibrate problem difficulty.
Simulated apps for agent training
A working copy of software built so an AI agent can operate it inside an RL environment.
05
Datasets
Open datasets and benchmarks from our published papers, each on its research page.
2026World Modeling for Agentic RL
A language model that stands in for an environment while an agent trains.
2026FigmaTrace
Human design workflows in Figma, captured as training data for creative design tasks.
2026SpeedrunBench
Video game speedrunning as a test of whether agents can beat strategies humans already know.
2025TRAIL
A benchmark for finding the step where an agent went wrong.
2025BLUR
573 tip-of-the-tongue questions that need search, reasoning and tool use.
2026TRACE
54 kinds of reward exploit in code RL, and how well models catch them.
2026GlobeBench
A benchmark for whether text world models keep track of state, rules and side effects.
2023SimpleSafetyTests
100 prompts for finding critical safety risks in language models.
2023FinanceBench
Open-book financial question answering about publicly traded companies.
06
Research and blog
Our papers and posts.
