Hydrogen-1.0 open-weight science model built on GLM-5.3 beats GPT-6-Astra
josephdviviano · x · 2026-11-02
The Hydrogen team introduced Hydrogen-1.0, billed as the first open-weight model aimed at scientific research: GLM-5.3 (@Zaiorg) post-trained on their internal scientific data, lab records, and discovery trajectories.
- Scores 38.3% on FrontierScience-Research vs GPT-6-Astra's 33.3%, and 93.0% on GPQA-Diamonds.
- The team argues science should be open, collaborative, and democratized rather than proprietary — only open weights let researchers inspect, adapt, and build on AI's rapid progress.
More from Models
- Frontier models flunk 1,000-variant DNA classification: refusals, token blowouts, encrypted reasoning — _lewtun · 2026-11-03
- Qwen Head Dispels Rumors: Qwen 4.0 and 5.0 Will Stay Open Source — ResearchCrafty1804 · 2026-11-03
- Plus Subscriber Complaints About OpenAI's Opaque Usage Limits to UK's CMA — YouHadTooMuchToThink · 2026-11-03
- OpenAI's Cyber Guardrails Softlock a User's Legit Security Audit Mid-Task — Linuxxsxx · 2026-11-03
- Anthropic S-1 shows revenue doubling to $16.8B/quarter; Gemini 4 Carbon takes #1 on new index — altryne · 2026-11-03
- Musk Throttles Grok Bot Limits Citing 'Unsustainably High' Costs — TechCrunch AI · 2026-11-03