Pushing frontier capabilities with RL and data.

Our goal is to address the largest capability gaps in frontier coding models. We solve this by creating tough and realistic environments with verifiable reward to help models improve at being able to find and judge ambiguous research-heavy solutions.

Quality is our strongest focus. The best data requires the best experts. Our experts come from places like Meta, Together AI, CMU and Berkeley, and the toughest QC in the business decides what ships.