patronus.ai is built with Webflow, hosted on Webflow Hosting behind Cloudflare. It measures traffic with Google Analytics and Google Tag Manager. It sends email through Mailchimp Transactional (Mandrill).
| Technology | Version | First seen |
|---|---|---|
| CMS & Site Builders1 | ||
| – | 7 Nov 2026 | |
| JavaScript Frameworks & Libraries1 | ||
| 3.5.1 | 7 Nov 2026 | |
| CDN & Hosting2 | ||
| – | 7 Nov 2026 | |
| Analytics & Tag Management2 | ||
| – | 7 Nov 2026 | |
| Transactional Email1 | ||
Registration from the registry's RDAP record, the certificate the site serves, and its rank in public traffic lists.
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
Counted from every stack change we've recorded: of the sites we watched running one of this site's tools, how many went on to add another. A pattern across sites, not a forecast for this one.
smtp.google.comns-1959.awsdns-52.co.ukns-566.awsdns-06.netns-343.awsdns-42.comns-1095.awsdns-08.orgMS=5B2A702D490F572EBA9A39A3F610F8AABC7487EBv=spf1 include:dc-aa8e722993._spfm.patronus.ai include:servers.mcsv.net ~allgoogle-site-verification=LU0gSCWJT700dcVjhnpel1FDd5r5SJibaTRtlxqIF48proxy-ssl.webflow.comgoogle-site-verification=hACeciLq3_esWpJRXWbqApWhzdJIaflYGQ8X2eoPfBkgoogle-site-verification=zCtGBHYQnT6_zOairj4dDt_QdxJhdwFf23JAFwpfGM0v=DMARC1; p=quarantine; pct=100; ri=86400; rua=mailto:dmarc@patronus.ai;User-agent: * Allow: / User-agent: GPTBot Allow: / User-agent: OAI-SearchBot Allow: / User-agent: ClaudeBot Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / User-agent: meta-externalagent Allow: / User-agent: meta-webindexer Allow: / Sitemap: https://www.patronus.ai/sitemap.xml Sitemap: https://www.patronus.ai/sitemap.xml
# Patronus AI > Patronus AI is a behavioral research lab for AI that studies where models fail and builds training data and simulated environments to improve them. Founded in 2023 by Anand Kannappan and Rebecca Qian, we work with frontier labs and enterprises. Our work covers and expands beyond five core domains: alignment, coding, knowledge work, science and computer use. ## About - [Home](https://www.patronus.ai/): What Patronus AI does and featured research ([Markdown](https://cdn.patronus.ai/site/md/index.md)) - [Company](https://www.patronus.ai/company): Founders, mission, backing and team ([Markdown](https://cdn.patronus.ai/site/md/company.md)) - [Domains](https://www.patronus.ai/domains): The five domains we build environments for ([Markdown](https://cdn.patronus.ai/site/md/domains.md)) - [Research](https://www.patronus.ai/research): All published papers ([Markdown](https://cdn.patronus.ai/site/md/research.md)) - [Blog](https://www.patronus.ai/blog): Research write-ups and engineering notes ([Markdown](https://cdn.patronus.ai/site/md/blog.md)) - [News](https://www.patronus.ai/news): Company announcements and press ([Markdown](https://cdn.patronus.ai/site/md/news.md)) - [Join](https://www.patronus.ai/join): Open roles ([Markdown](https://cdn.patronus.ai/site/md/join.md)) - [Contact](https://www.patronus.ai/contact): Talk to the team. To sell or license data, go to https://data.patronus.ai/ or write to data@patronus.ai ([Markdown](https://cdn.patronus.ai/site/md/contact.md)) ## Solutions - [Post-training data](https://www.patronus.ai/post-training-data): RL environments and pairwise reviews of agent transcripts, measured by training open models on held-out tasks ([Markdown](https://cdn.patronus.ai/site/md/post-training-data.md)) ## Domains - [Alignment](https://www.patronus.ai/domains/alignment) ([Markdown](https://cdn.patronus.ai/site/md/domains/alignment.md)) - [Coding](https://www.patronus.ai/domains/coding) ([Markdown](https://cdn.patronus.ai/site/md/domains/coding.md)) - [Knowledge Work](https://www.patronus.ai/domains/knowledge-work) ([Markdown](https://cdn.patronus.ai/site/md/domains/knowledge-work.md)) - [Science](https://www.patronus.ai/domains/science) ([Markdown](https://cdn.patronus.ai/site/md/domains/science.md)) - [Computer Use](https://www.patronus.ai/domains/computer-use) ([Markdown](https://cdn.patronus.ai/site/md/domains/computer-use.md)) ## Research - [Few Word Do Trick](https://www.patronus.ai/blog/few-word-do-trick): Blog post on how three open model families cut filler words from their reasoning by a quarter to a half in 2026, and why shorter wording did not mean fewer tokens ([Markdown](https://cdn.patronus.ai/site/md/blog/few-word-do-trick.md)) - [GlobeBench: Evaluating Text-Based World Model Capabilities for Scalable Environment Simulations](https://www.patronus.ai/research/globebench) ([Markdown](https://cdn.patronus.ai/site/md/research/globebench.md)) - [Introducing GlobeBench](https://www.patronus.ai/blog/globebench): Blog post on GlobeBench, which tests whether language models can stand in for an environment by predicting its response to an agent's next action ([Markdown](https://cdn.patronus.ai/site/md/blog/globebench.md)) - [Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL](https://www.patronus.ai/research/world-modeling-for-agentic-rl) ([Markdown](https://cdn.patronus.ai/site/md/research/world-modeling-for-agentic-rl.md)) - [FigmaTrace: Capturing Creative Nuances in Human Figma Design Workflows](https://www.patronus.ai/research/figmatrace) ([Markdown](https://cdn.patronus.ai/site/md/research/figmatrace.md)) - [SpeedrunBench: Challenging LLM Agents with Video Game Speedrunning](https://www.patronus.ai/research/speedrunbench) ([Markdown](https://cdn.patronus.ai/site/md/research/speedrunbench.md)) - [Adaptive Environments For LLM Agents Need to Jointly Scale Task, Scaffolding and Reward Design](https://www.patronus.ai/research/generative-simulators) ([Markdown](https://cdn.patronus.ai/site/md/research/generative-simulators.md)) - [Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis](https://www.patronus.ai/research/trace) ([Markdown](https://cdn.patronus.ai/site/md/research/trace.md)) - [DETOUR: An Interactive Benchmark for Dual-Agent Search and Reasoning](https://www.patronus.ai/research/detour) ([Markdown](https://cdn.patronus.ai/site/md/research/detour.md)) - [HalluWorld: A Controlled Benchmark for Hallucination via Reference World Models](https://www.patronus.ai/research/halluworld) ([Markdown](https://cdn.patronus.ai/site/md/research/halluworld.md)) - [Unlocking Prompt Infilling Capability for Diffusion Language Models](https://www.patronus.ai/research/prompt-infilling-for-diffusion-lms) ([Markdown](https://cdn.patronus.ai/site/md/research/prompt-infilling-for-diffusion-lms.md)) - [TRAIL: Trace Reasoning and Agentic Issue Localization](https://www.patronus.ai/research/trail) ([Markdown](https://cdn.patronus.ai/site/md/research/trail.md)) - [Browsing Lost Unformed Recollections: A Benchmark for Tip-of-the-Tongue Search and Reasoning](https://www.patronus.ai/research/blur) ([Markdown](https://cdn.patronus.ai/site/md/research/blur.md)) - [MEMTRACK: Evaluating Long-Term Memory and State Tracking in Multi-Platform Dynamic Agent Environments](https://www.patronus.ai/research/memtrack) ([Markdown](https://cdn.patronus.ai/site/md/research/memtrack.md)) - [Humanity’s Last Exam](https://www.patronus.ai/research/humanitys-last-exam) ([Markdown](https://cdn.patronus.ai/site/md/research/humanitys-last-exam.md)) - [Contrastive Decoding Mitigates Score Range Bias in LLM-as-a-Judge](https://www.patronus.ai/research/contrastive-decoding-for-llm-judges) ([Markdown](https://cdn.patronus.ai/site/md/research/contrastive-decoding-for-llm-judges.md)) - [Position: We Need A Unified Definition of Hallucination (It’s The World Model, Stupid!)](https://www.patronus.ai/research/unified-hallucination-framework) ([Markdown](https://cdn.patronus.ai/site/md/research/unified-hallucination-framework.md)) - [GLIDER: Grading LLM Interactions and Decisions using Explainable Ranking](https://www.patronus.ai/research/glider-sota-slm-judge) ([Markdown](https://cdn.patronus.ai/site/md/research/glider-sota-slm-judge.md)) - [Lynx: An Open Source Hallucination Evaluation Model](https://www.patronus.ai/research/lynx-hallucination-detection-model) ([Markdown](https://cdn.patronus.ai/site/md/research/lynx-hallucination-detection-model.md)) - [FinanceBench: A New Benchmark for Financial Question Answering](https://www.patronus.ai/research/financebench) ([Markdown](https://cdn.patronus.ai/site/md/research/financebench.md)) - [SimpleSafetyTests: a Test Suite for Identifying Critical Safety Risks in Large Language Models](https://www.patronus.ai/research/simplesafetytests) ([Markdown](https://cdn.patronus.ai/site/md/research/simplesafetytests.md)) ## Guides - [Guide to RL environments](https://www.patronus.ai/guide-to-rl-environments) ([Markdown](https://cdn.patronus.ai/site/md/guide-to-rl-environments.md)) - [AI world models](https://www.patronus.ai/ai-world-models) ([Markdown](https://cdn.patronus.ai/site/md/ai-world-models.md)) - [AI agent development](https://www.patronus.ai/ai-agent-development) ([Markdown](https://cdn.patronus.ai/site/md/ai-agent-development.md)) - [LLM testing](https://www.patronus.ai/llm-testing) ([Markdown](https://cdn.patronus.ai/site/md/llm-testing.md)) - [AI reliability](https://www.patronus.ai/ai-reliability) ([Markdown](https://cdn.patronus.ai/site/md/ai-reliability.md)) ## Resources - [Resources](https://www.patronus.ai/resources): One catalog of our guides, comparisons, glossary, open datasets and research ([Markdown](https://cdn.patronus.ai/site/md/resources.md)) - [Best RL Environment Companies in 2026](https://www.patronus.ai/resources/rl-environment-companies): Nine RL environment companies compared, with what each sells and who it suits ([Markdown](https://cdn.patronus.ai/site/md/resources/rl-environment-companies.md)) - [Best Scale AI Alternatives for RL Environments in 2026](https://www.patronus.ai/resources/scale-ai-alternatives): Nine Scale AI alternatives for RL environments, with what each sells and who it suits ([Markdown](https://cdn.patronus.ai/site/md/resources/scale-ai-alternatives.md)) - [Best RL Environment Companies for Computer Use Agents in 2026](https://www.patronus.ai/resources/computer-use-rl-environment-companies): Eight companies that build RL environments for computer use agents, from app replicas and desktop VMs to live browsers, compared ([Markdown](https://cdn.patronus.ai/site/md/resources/computer-use-rl-environment-companies.md)) - [Best Financial LLM Benchmarks in 2026](https://www.patronus.ai/resources/financial-llm-benchmarks): Nine benchmarks for testing LLMs and agents on financial work, from filings QA and table reasoning to agentic research, compared ([Markdown](https://cdn.patronus.ai/site/md/resources/financial-llm-benchmarks.md)) - [RL Environment Startups to Watch in 2026](https://www.patronus.ai/resources/rl-environment-startups): Fourteen RL environment startups grouped by what they build, from research labs to software replicas and open hubs, and who is hiring ([Markdown](https://cdn.patronus.ai/site/md/resources/rl-environment-startups.md)) - [RL environment and post-training glossary](https://www.patronus.ai/glossary): Definitions of RL environment and post-training terms, from verifiers and pass@k to reward hacking ([Markdown](https://cdn.patronus.ai/site/md/glossary.md)) - [What is a verifier in RL?](https://www.patronus.ai/glossary/verifier): The check that decides whether an agent's answer or final state is correct, how verifiers fail and how to test one ([Markdown](https://cdn.patronus.ai/site/md/glossary/verifier.md)) - [Pass@k and RL difficulty calibration](https://www.patronus.ai/glossary/pass-at-k): The chance that at least one of k attempts succeeds, and how to use it to calibrate RL problem difficulty ([Markdown](https://cdn.patronus.ai/site/md/glossary/pass-at-k.md)) - [Simulated apps for agent training](https://www.patronus.ai/glossary/simulated-apps): Working copies of software built so an agent can operate them inside an RL environment ([Markdown](https://cdn.patronus.ai/site/md/glossary/simulated-apps.md)) ## Optional - [Full text of these pages](https://cdn.patronus.ai/site/md/llms-full.txt): every page above in one Markdown file
| – |
| 7 Nov 2026 |
| – |
| 7 Nov 2026 |
| – |
| 7 Nov 2026 |