Deep Principle releases Hydrogen-1.0, a 753B open-weight model for chemistry, materials and life sciences
The Hangzhou company post-trained Z.ai's GLM-5.3 on lab records and discovery trajectories and reports 38.3% on FrontierScience-Research against 33.3% for GPT-6-Astra; the scores are self-reported and the training data is not released.
Why it matters
Most recent AI-for-science results have come from closed models, and Deep Principle pitches Hydrogen-1.0 as an inspectable alternative under the MIT License that researchers can run and audit themselves. Its reported lead is narrow and comes from the company's own runs, and it trails GPT-6-Astra on NMR structure elucidation. Watch for independent replication of the benchmark claims and whether labs adopt it for experiment planning, where the company reports it beat Kimi-K3.
Key facts
- 753B parameters, post-trained from GLM-5.3
- 38.3% on FrontierScience-Research vs 33.3% for GPT-6-Astra (self-reported)
- 93.0% on GPQA-Diamond (reported)
- Released under the MIT License
- 1M-token context, 128K maximum output
- Full datasets and recipes not released