Hillclimb, Inc.

Frontier AI labs need large volumes of high-quality research data and reinforcement learning environments to push model capabilities toward recursive self-improvement, but creating these environments at scale and aggregating human research data are both difficult, labor-intensive problems. Hillclimb frames this as two core bottlenecks: aggregating all human research data and scaling RL environment creation. Without solving these, labs are limited in how much stronger research capability can emerge from their models. The company positions itself as addressing this gap directly for the labs that need it.

Hillclimb creates RL environments and sells them to frontier labs to help improve model capabilities toward recursive self-improvement. Its stated approach is to make incremental progress on aggregating human research data and scaling environment creation, with each step unlocking an order-of-magnitude increase in available environments. The company points to a previous venture as a model for this method: it built a community of IMO medalists who co-trained a state-of-the-art math model together with Nous Research. The site positions this same community-driven, domain-expert approach as the template for scaling environment creation for AI research more broadly.