Ulam

Ulam sells reasoning datasets, expert-reviewed traces, benchmarks, and verifier-backed RL environments—primarily for mathematics and STEM—to AI labs and developers.

Source:ulam.ai· Sep 29, 2026

ulam.ai · updated Oct 13, 2026

RL environments Training data Evals and benchmarks Human data

Add to compareSuggest an edit

Headquarters: Warsaw, Poland
Source:linkedin.com· Sep 29, 2026
Founded: 2017
Source:linkedin.com· Sep 29, 2026
Headcount: 1-10
Source:linkedin.com· Sep 29, 2026
Total funding: Not publicly listed
Latest round: Not publicly listed

Trust & data

How the vendor protects buyer material, where its data comes from, and where it can run.

Data sources: Synthetic / generated
Source:ulam.ai· Sep 29, 2026
Quote
Synthetic / generated: We are building a low-contamination math benchmark from synthetic Erdős-style research problems and IMO-like proof problems
Stated confidentiality practices: Ulam says private splits, hidden verifier keys, review materials, manifests, and answer keys stay sealed or protected; it does not specify access-control or NDA practices.
Sources:ulam.ai/erdosbench.html ulam.ai/data· Sep 29, 2026
Not publicly listed: SOC 2, Trust center, Other certifications, Deployment, Exclusivity

What they sell

Offerings: RL environments Training data Evals and benchmarks Human data
Sources:ulam.ai (1)ulam.ai/data· Sep 29, 2026
Domains: Coding Science Math Multi-domain
Sources:ulam.ai (1)ulam.ai/blog
Delivery formats: DatasetsAPI
Source:ulam.ai· Sep 29, 2026
Public samples & benchmarks: Verified Research Reasoning
OlympiadNet-Math
AIME++
Math RL Tasks
SOTA Math
ErdősBench
SimoBench
Source:ulam.ai· Sep 29, 2026
Not publicly listed: Named customers

Every change to these facts is kept in the public claims log. Record co_bb97e342c467.