Hillclimb

Hillclimb creates custom mathematics-focused reinforcement-learning environments and supervised fine-tuning data from expert problem-solving, selling them to frontier AI labs to train models and research agents.

Sources:hillclimb.com ycombinator.com

hillclimb.com · updated Sep 29, 2026

RL environments Human data Training data

Add to compareSuggest an edit

Headquarters: San Francisco, United States
Source:ycombinator.com
Founded: 2025
Source:ycombinator.com
Headcount: 1-10
Source:ycombinator.com
Total funding: $500K
Source:vcbacked.co
Latest round: Pre-Seed, $500K, December 2025
Source:vcbacked.co

Trust & data

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

Data sources: Expert or contractor-created
Source:data.hillclimb.ing
Quote
Expert or contractor-created: Curated set of 50 original mathematical problems designed by Hillclimb researchers
Stated confidentiality practices: Hillclimb says it vets talent; no specific NDAs, access controls, or customer-data isolation practices are stated.
Source:ycombinator.com
Not publicly listed: SOC 2, Trust center, Other certifications, Deployment, Exclusivity

What they sell

Offerings: RL environments Human data Training data
Sources:hillclimb.com ycombinator.com
Domains: Math
Source:hillclimb.com
Delivery formats: Datasets
Source:data.hillclimb.ing
Public samples & benchmarks: Hillclimb Benchmark v1
Putnam 2024 Grading Rubric
Generic Rubric v1
Generic USAMO Style Rubric
Math Rubrics Collection
Nomos Corpus Preview
Sources:data.hillclimb.ing (1)data.hillclimb.ing/putnam-rubric data.hillclimb.ing/generic-rubric data.hillclimb.ing/usamo-rubric
Named customers: Nous Research
Source:ycombinator.com

Every change to these facts is kept in the public claims log. Record co_48d8fa9c0cf4.