Hydrogen 1.0 science model post-trained from GLM 5.3 for scientific discovery
Hydrogen 1.0 is a model post-trained from GLM 5.3 for scientific work, scoring 38.33% on Frontier Science at 25 cents per problem.
Hydrogen 1.0 is a model post-trained from GLM 5.3, an open-weights model from a Chinese lab, for scientific work. Matthew Berman attributed it to MIT and said it scores 38.33% on the Frontier Science benchmark at 25 cents per problem, against 88 cents per problem for GPT-6 Astra. Post-training means taking a model that has already been trained and training it further on material aimed at a particular kind of task. In this case the stated aim was scientific discovery, and the base model is the open-weights GLM 5.3.
Berman read the benchmark chart as showing Hydrogen 1.0 ahead of GLM 5.3, Kimi K3 and GPT-6 Astra, while costing less than all of them except GLM 5.3. The walkthrough reads only Hydrogen's 38.33% aloud and doesn't give the other models' scores, so the size of the lead can't be judged from it. It also doesn't explain how the per-problem price is measured.
The demo Berman showed was a simulation of a polyethylene melt in 3D at the molecular level, with the model writing the code. The rendered molecules can be rotated on screen, and the readouts he named were density, RDF, radius of gyration and end-to-end distance. Radius of gyration and end-to-end distance describe how large and how stretched the polymer chains are. RDF, the radial distribution function, describes how neighboring atoms are spaced. The segment shows the code running and the outputs being displayed, but it doesn't say whether those results were checked against known values or reference simulations.
Berman said the weights are on Hugging Face and that the model can also be used through an API from a provider. The name of that provider is unclear in the auto-generated captions of the video. Beyond the benchmark score, the price comparison and the single polyethylene demo, the walkthrough offers no other tasks, results or limits for Hydrogen 1.0.
What changed
Hydrogen 1.0, a model post-trained from the open-weights GLM 5.3 for scientific discovery, is available.
Type: New capability · Applies to: Hydrogen 1.0
Why it matters
Cheaper domain-specialised science models could make computational research workflows affordable for smaller labs and companies.
What you can do
R&D teams could check whether the benchmark cost advantage holds on their own scientific tasks before switching.
Original source
We have no link to the original source for this story yet.
Where we saw it
Forward Future (Matthew Berman)
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works