Deep Principle has launched Hydrogen-1.0, a 753 billion parameter open-weight model designed for scientific discovery in fields like chemistry and life sciences. This model is a derivative of GLM-5.3, enhanced with domain-specific data, laboratory records, and agent discovery trajectories. Hydrogen-1.0 reportedly outperforms GPT-6-Astra on the FrontierScience-Research benchmark and demonstrates capabilities in tasks such as chemical reaction planning and data analysis, with weights available under the MIT License. AI
IMPACT This model's open-weight nature and scientific focus could accelerate reproducible research and discovery in specialized fields.
RANK_REASON Frontier-lab model release with system card and benchmark results.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →