What ulam.ai does

ulam.ai (legally Yotta Content LTD, registered in London, UK) is an AI research lab that builds research-grade reasoning datasets and evaluation systems for frontier AI labs and AI developers. The core technology combines LLM-guided reasoning with Lean 4 formal verification through UlamAI Prover, an open-source workflow that converts natural-language mathematical arguments into machine-checkable proofs. The portfolio covers four product layers: (1) core evaluation and training infrastructure (UlamAI Prover, Ulam Bench for private benchmarks, Ulam Arena for agent evaluation, UlamGym for RLVR reward environments); (2) proprietary data products built from Erdős-style, arXiv, and olympiad sources (including the 1,146-problem UnsolvedMath corpus, 20,000+ OlympiadNet-Math problems, and 1,000+ research-level trajectories); (3) benchmark products (ErdosBench, SimoBench); and (4) inference-economics research (RAVE/QeRAVE MoE compression) plus a consumer-facing Ulam Alloy API for model routing.

The company monetizes via subscription/recurring revenue streams: Ulam Data (verified datasets, expert-reviewed trajectories, grader packs), private benchmark programs (monthly rounds with hidden holdouts and grader-backed scorecards), UlamGym RLVR access, and Ulam Alloy API tiers. Go-to-market is hybrid — direct enterprise sales to frontier AI labs using a scoping/pilot/expansion engagement model (2-3 week scoping, 10-12 week pilot, recurring expansion), combined with open-source distribution via GitHub and Hugging Face for community-driven adoption. Primary customers are frontier AI labs and AI developers needing harder-to-game signals on mathematical reasoning beyond solved public benchmarks, with secondary segments in RL/training teams and academic reasoning researchers. The company operates from EU+US with no disclosed funding rounds and appears founder-controlled under CEO Przemek Chojecki.

Name

ulam.ai

Legal name

Yotta Content LTD

Website

https://ulam.ai

Company type

Private

Founded year

2017

Operating status

Operating

Headcount range

1–10 employees

Short description

ulam.ai builds research-grade reasoning datasets, Lean 4-verified evaluation systems, and inference-economics research for frontier AI labs and AI developers, monetizing through subscriptions for private benchmarks, RLVR training environments, and custom data products.

Ownership category

akta.pro rank

Product category

AI Mathematical Reasoning Evaluation Software

NAICS

Custom Computer Programming Services (541511)

SIC

Services-Computer Programming Services (7371)

akta.pro primary industry

AI Observability, Monitoring & Evaluation Platforms (Drift, Quality, Safety) (HDAEANAF)

akta.pro secondary industries

Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC), Audit, Explainability & Accountability Tooling (traceability, reporting) (HDAAAKAL)

Keywords

• Mathematical AI benchmarks
• Lean 4 formal verification
• RLVR training datasets
• AI reasoning datasets
• AI model evaluation

Headquarters

HQ city

Warsaw

HQ country

Poland

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations

Revenue model

1. Ulam Data Products: Sale of verified datasets, expert-reviewed trajectories, proof attempts, grader packs, and failure-to-data conversion with one month of support to help teams develop stronger verified reasoning capabilities.
2. Private Benchmark Programs: Custom benchmark suites and eval protocols for AI labs requiring private, difficult-to-overfit measures of frontier reasoning. Includes monthly private rounds with hidden problems, proof-validity judging, and technical readouts.
3. UlamGym RLVR Access: Verifier-only reward API access with prompt/export bundles, hidden manifests, redacted trainer replays, reward vectors, and monthly RLVR task refresh from benchmark failures.
4. Ulam Alloy API: AI routing product with subscription tiers for API access and model routing services. Includes free tier with limited messaging.

Pricing tiers

Go-to-market motion3 records

Distribution channels4 records

Marketing channels6 records

Core offering

ulam.ai is a mathematical AI research lab building reasoning datasets, formal-verification tooling, and benchmarking infrastructure. Its core products include the UlamAI Prover for Lean 4 proof automation, the Ulam Bench and Ulam Arena evaluation/benchmarking suites, UlamGym as an RLVR training environment, and curated reasoning datasets such as OlympiadNet-Math, UnsolvedMath, ErdosBench, and SimoBench. These offerings are aimed at AI labs and reasoning-research teams seeking training data, evaluation tooling, and proof-automation models.

Product overview

ulam.ai is an AI research lab focused on reasoning data and evaluation systems for frontier AI models, with mathematics as the primary test case. The portfolio consists of: (1) core products for mathematical reasoning - UlamAI Prover (open-source Lean 4 verifier), Ulam Bench (private evaluation benchmarks), Ulam Arena (agent evaluation environments), and UlamGym (RLVR reward environments); (2) data products including Research-Level Reasoning Datasets, OlympiadNet-Math (20,000+ problems), UnsolvedMath (1,146 open problems), and Custom SFT/RLVR datasets; (3) benchmark products including ErdosBench (226 problems) and SimoBench (126 synthetic IMO problems); (4) Ulam Alloy - an AI routing API/chat interface; and (5) RAVE - a MoE compression technology. The core workflow connects data creation, private benchmarking, and verifier-backed RLVR training loops.

Differentiator

Problem solved

Functional benefit

Products and services

• UlamAI Prover A prover built around Lean 4 formal verification, intended for AI labs and reasoning-research teams that need automated proof tooling.
• Ulam Bench An evaluation/benchmarking suite designed to measure the mathematical reasoning performance of AI models.
• Ulam Arena An evaluation environment for head-to-head comparison of mathematical reasoning models.
• UlamGym A training environment built for RLVR-style mathematical reasoning model training.
• OlympiadNet-Math A curated olympiad-level mathematical problem dataset intended for reasoning-model training.
• UnsolvedMath A dataset of unsolved mathematical problems used to evaluate reasoning models on research-level mathematics.
• ErdosBench A benchmark dataset evoking Erdős-style combinatorial problems for mathematical-reasoning model evaluation.
• SimoBench A benchmark for mathematical simulation-style reasoning evaluation.
• Ulam Alloy An Alloy-style reasoning offering tied to the ulam.ai product line.
• RAVE AI model architecture (Router-Aware Virtual Experts) with QeRAVE quantization, sold/deployed as a model offering.

Quantifiable outcome

• Full solution to Erdos Problem 1148 with Lean 4 formalization verified in March 2026
• +2 more outcomes

Segments3 records

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

AI capability9 records

Feature6 records

Scale indicators8 records

Recent moves6 records

Expansion highlights5 records

Others

• Hugging Face: Hugging Face is both a distribution partner (Ulam hosts datasets on HF) and an adjacent platform offering Spaces, Evaluation, and dataset services that overlap with Ulam's data and benchmark distribution model.
• Lean FRO / Mathlib community: The Lean theorem prover community (Lean FRO, Mathlib) is the open-source ecosystem underpinning UlamAI Prover. While not a competitor, it is the enabling technology whose roadmap and toolchain shape Ulam's verification capability.
• Weights & Biases: W&B provides ML experiment tracking, model evaluation, and observability tooling. Its evaluation and LLM-judge capabilities are adjacent to Ulam Bench and Ulam Arena for ML teams that need to track reasoning-model performance over time.
• Epoch AI: Epoch AI is a research org that benchmarks frontier models and tracks AI capability trajectories, including mathematical reasoning. Comparable to Ulam's ErdosBench and SimoBench as a third-party eval/benchmark provider.

Broad incumbents

• Scale AI: Scale AI is a broad incumbent providing data labeling, RLHF, and AI evaluation services to frontier labs. While Scale's eval breadth is much wider than math, its SEAL evaluation suite and private benchmarking compete for the same frontier-lab eval budget.
• DeepMind (AlphaProof / AlphaGeometry): DeepMind's AlphaProof combines reinforcement learning with Lean for formal mathematical reasoning and is a well-funded, broad incumbent pursuing the same AI-for-math frontier that Ulam targets with UlamAI Prover and Erdős solutions.
• Surge AI: Surge AI provides high-quality RLHF data and evaluation services for frontier AI labs. It competes with Ulam for the same reasoning-data and human-in-the-loop evaluation budget at frontier model developers.

Direct peers

• Numina AI: Numina builds high-quality mathematical reasoning datasets (NuminaMath, MATH dataset lineage) and tools for training reasoning models. It is the closest direct peer to Ulam's research-grade math data products, both targeting post-training of frontier reasoning models.
• Harmonic AI: Harmonic AI is building mathematical superintelligence with formal verification (Lean), aiming to solve open mathematical problems using AI. Directly comparable to Ulam's Erdős-focused, Lean-verified approach to mathematical reasoning.

Emerging players

• Symbolica AI: Symbolica AI is an emerging player building structured reasoning and mathematical AI, focused on formal methods and interpretable architectures. Comparable to Ulam's research-first, math-grounded approach to AI reasoning.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks4 records

Key highlights6 records

Customer concentration

• GitHubgithub.com/ulamai
• LinkedInlinkedin.com/company/ulam-ai

Revenue estimate

Valuation estimate

Number of profiles

Profiles1 record

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about ulam.ai