Deep Principle releases open-weight Hydrogen-1.0 and says it beats GPT-6-Astra at science
Co-founder Chenru Duan says the GLM-5.3 post-train, trained on Deep Principle's own lab records, scores 38.3% on FrontierScience-Research against GPT-6-Astra's 33.3%.
By Ryan Merket · Published
Primary source: x.com
Why it matters
Hydrogen-1.0 tests whether a materials startup's own lab records and discovery trajectories are a training advantage that general labs cannot copy from public data. Because the weights are open and priced by the company at GLM-5.3 rates, other researchers can rerun the FrontierScience-Research and GPQA-Diamond claims and judge whether post-training on full experiment cycles produces better scientific decisions, not better answers.
Deep Principle co-founder and CTO Chenru Duan (@chenru_duan) introduced Hydrogen-1.0 on November 2nd. It is the Hangzhou startup's first open-weight model for scientific research, built by post-training Z.ai's GLM-5.3 on the company's internal scientific data, laboratory records and discovery trajectories.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.