The Deep Principle team has launched Hydrogen-1.0, a specialized AI model designed to automate scientific discovery. Built as an open-weight system post-trained from GLM-5.3, the model was developed using the company’s internal lab records and discovery trajectories to assist researchers in inspecting and adapting the tool for their own work.
In performance tests, the model achieved a 38.3% score on the FrontierScience-Research benchmark, surpassing the 33.3% posted by GPT-6-Astra, and recorded a 93.0% score on GPQA-Diamonds. The developers stated the release is intended to democratize scientific research by providing an open alternative to proprietary superintelligence models.
