Hydrogen-1.0: an open-weight science model that its maker says beats GPT-6 Astra on research questions

Deep Principle released Hydrogen-1.0, built by training GLM-5.3 further on chemistry, materials and life-sciences data plus records from its own labs. It says the model scores 38.3% on FrontierScience-Research against 33.3% for GPT-6 Astra, and its demos show it writing and fixing a simulation workflow and planning lab experiments; these are the company's own tests. The weights are free on Hugging Face under the MIT licence, and there's an API through NeuroSpark.

Why it mattersRelevant if you do R&D in chemistry or materials: a free model you can run privately on your own data. It's very large (around 753 billion parameters, per MarkTechPost), so most teams will use the API rather than host it.

Curated by Sara Simeone as part of our human editorial coverage. The original sources are linked below; curation is distinct from a Hype or Fact? assessment.

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.