FRONTIER LAB / LLM · WORTH A LOOK
Hydrogen-1.0 (Deep Principle)
Open-weight science model from Deep Principle, post-trained from GLM-5.3 for chemistry, materials and life sciences.
What Hydrogen-1.0 (Deep Principle) does
Deep Principle's Hydrogen-1.0 is GLM-5.3 trained further on chemistry, materials-science and life-sciences data, plus records and agent trajectories from the company's own labs. It's aimed at research work: writing and fixing simulation workflows, predicting reactions, reading NMR spectra and planning experiments. Weights are on Hugging Face and ModelScope under the MIT licence; there's also an API via NeuroSpark and Deep Principle's Mira 'AI Scientist'.
Good for
- R&D teams in chemistry and materials science
- Drafting experiment plans and lab protocols
- Scripting and debugging simulation workflows
Try this in 10 minutes
Honest take
No red flags in our notes so far
- All benchmark results (e.g. 38.3% vs 33.3% for GPT-6 Astra on FrontierScience-Research) are Deep Principle's own evaluations and haven't been independently replicated.
- It's a very large model; most small teams will use the API rather than host it.
Pricing
Open weights, free (MIT licence). API via NeuroSpark (priced like GLM-5.3, per Deep Principle). Self-hosting needs data-centre GPUs (about 753B parameters, per MarkTechPost).
Prices as listed by the maker when we last checked (2026-11-03); usually in US dollars. Check the website for current prices.
Latest update
- : Hydrogen-1.0 released with open weights (MIT); Deep Principle says it beats GPT-6 Astra on FrontierScience-Research in its own tests.