Math RL Environments for Post-training
We work with the world's most brilliant mathematicians to design RL environments that enable LLMs to achieve state-of-the-art model performance.
Key people at hillclimb.
hillclimb was founded in 2025 by Jun Park (Founder) and Ibrakhim Ustelbay (Founder).
Hillclimb is a cutting-edge company focused on designing mathematics-focused reinforcement learning (RL) environments to accelerate post-training of foundational AI models, particularly targeting the development of artificial general intelligence (AGI) and superintelligence. They build their product by collaborating with a dense community of elite mathematicians, including IMO medalists and Putnam winners, to generate high-quality, rigorously verified math data that serves frontier AI research labs. This approach solves the problem of slow, low-quality training data delivery by enabling rapid, iterative feedback loops directly between researchers and top math talent, significantly improving data quality and research velocity[1][2].
Founded in 2025 by Jun Park and Ibrakhim Ustelbay, Hillclimb emerged from their shared background in competitive esports and AI research, including experience at DeepMind and AI interviewing platforms. Their mission is to close the gap between current AI capabilities and the arrival of artificial superintelligence by leveraging the smartest human minds to create the next generation of training data. Early traction includes successful partnerships with leading AI labs like Nous, where their mathematicians receive co-authorship for their critical contributions to AI research datasets[1].
Hillclimb rides the critical trend of improving training data quality and sophistication to enable leaps in AI capability, particularly in reinforcement learning and AGI development. As AI models like GPT-5 and beyond increasingly assist in coding, research, and scientific discovery, the demand for specialized, high-quality training environments grows. Hillclimb’s timing is pivotal as the AI field transitions from large-scale pretraining to post-training fine-tuning and reinforcement learning with human-level expertise embedded in the data. Their work influences the broader ecosystem by setting new standards for data quality, collaboration, and speed, helping frontier labs push the boundaries of AI research[1][2][3].
Hillclimb is poised to become a foundational player in the AI training data ecosystem, especially as the race toward superintelligence intensifies. Future growth will likely involve expanding their talent network, deepening partnerships with leading AI labs, and possibly broadening their RL environment offerings beyond mathematics. Trends such as increasing AI model complexity, the rise of open-source RL environments, and the growing importance of human-in-the-loop data generation will shape their trajectory. Their influence may evolve from a niche math data provider to a critical infrastructure partner for next-generation AI development, helping to accelerate the arrival of autonomous scientific discovery and artificial superintelligence[1][2][3].
Key people at hillclimb.
hillclimb was founded in 2025 by Jun Park (Founder) and Ibrakhim Ustelbay (Founder).
