Hydrogen-1.0: Deep Principle’s AI model for materials discovery

Yahoo Finance（PR Newswire） USA

Overview

Deep Principle has released Hydrogen-1.0, an open-weight scientific model designed to automate experimental planning and materials discovery. The model outperforms GPT-6-Astra and Kimi-K3 in specific transition metal complex design and scientific computing tasks. This integration of AI with laboratory workflows aims to reduce the number of attempts required to find materials with desired properties.

In Depth

Key Findings

Hydrogen-1.0 demonstrates superior performance in iterative materials discovery compared to other large language models. In a transition metal complex design evaluation seeking a larger HOMO–LUMO gap, Hydrogen-1.0 achieved a mean best-so-far gap of 4.318 eV by the fifth round, surpassing GPT-6-Astra (3.689 eV) and Kimi-K3 (3.353 eV). Additionally, in internal scientific computing and one-step forward reaction prediction, Hydrogen-1.0 outperformed GPT-6-Astra across all thinking settings, leading by 5.13 and 4.76 percentage points against Astra’s maximum thinking setting.

Technical Details

Built via post-training of GLM-5.3, Hydrogen-1.0 utilizes chemistry, materials science, and life sciences data, including agent trajectories from Deep Principle’s laboratories. The training process employs supervised fine-tuning for reasoning and experimental protocols, while reinforcement learning incorporates feedback from research outcomes. The model is capable of:

• Experiment Design: Converting research goals into JSON protocols. In a 12-vial Suzuki coupling example, the model detected and corrected an aspiration error during a virtual rehearsal.
• Simulation Workflow: Writing Python to build a polyethylene melt and running GROMACS, including the ability to repair topology errors.
• Pathway Exploration: Generating 27 molecular structures via 30 reactions in an Ireland–Claisen rearrangement network to identify bottlenecks.

Background & Context

The model is accessible via open-weight downloads on Hugging Face and ModelScope, an API on NeuroSpark, or through Mira, Deep Principle’s AI Scientist. Development involved collaborations with UniPat and Mind Lab on data and training pipelines, and NeuroSpark on infrastructure. In benchmark evaluations, Hydrogen-1.0 outperforms GPT-6-Astra on FrontierScience-Research and improves upon GLM-5.3 on GPQA Diamond.

Significance & Outlook

By training on complete discovery trajectories—connecting desired properties to synthesis and evaluation—Hydrogen-1.0 enables researchers to prioritize investigations more effectively. Deep Principle intends to extend this learning across different material families so that experience from one campaign improves subsequent discovery efforts.

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.

• Your email and selected fields are used only to deliver the newsletter.
• We never share your information with third parties.
• You can unsubscribe anytime via the link in each email.

See our Privacy Policy for details.

Agree & Continue ▶

Takes about a minute · Unsubscribe anytime