Deep Cogito

Deep Cogito is a post-training research lab working on the stage after a model has been pre-trained, applying large-scale reinforcement learning and a method it calls Iterated Distillation and Amplification. That method spends extra computation to reach better answers, then trains those gains back into the model weights so the improvement becomes permanent.

The lab demonstrates the approach publicly through its Cogito family of open-weight models, published on Hugging Face and served by inference providers including Together AI, Fireworks AI, and Baseten. It sells the same engine to companies that want a specialized model trained on their own data, decisions, and outcomes, so the customer owns the resulting model rather than renting a general-purpose one.