Deep Principle's Hydrogen-1.0 beats GPT-6 Astra on scientific benchmarks
卡尔的AI沃茨 · wechat · 2026-11-11
AI4S startup Deep Principle released Hydrogen-1.0, post-trained from GLM-5.3 on scientific knowledge and full discovery trajectories via SFT+RL. It scores 38.3% on FrontierScienceResearch vs GPT-6 Astra's 33.3% at lower API cost, with open weights. Hands-on tests show active self-verification: it discards statistically unsupported conclusions and ships reproducible scripts; official demos span GROMACS simulation, reaction networks, NMR structure solving, experiment planning (5.59 avg vs Kimi K3's 4.57) and 5-round TMC material search (4.318eV vs Astra's 3.689eV).
More from Models
- Open-weights Inkling Small 2 impresses: researcher says he prefers it over DeepSeek V4.1 Flash — gnukeith · 2026-11-13
- Manual Review Exposes How LLMs Cheat Benchmarks: Sandbox Escapes and Copying Peers — teortaxesTex · 2026-11-13
- Models invent their own filler tokens during RL training — code_star · 2026-11-13
- Google Search reportedly gets massive Gemini 4 Hydrogen upgrade with in-search simulations — AI_Andrew · 2026-11-13
- Google ships Gemini 4 Hydrogen: beats its bigger Argon model on intelligence index, 3x coding gains — 量子位 · 2026-11-13
- Chinese medical AI beats 250 top doctors 88-12 on complex lung cancer cases in new study — 量子位 · 2026-11-13