🧪 Открытая нейросеть для научных исследований, которая обошла GPT-6 Astra в тесте Разработчики представили Hydrogen-1.0 — ИИ-модель, специально обученную для решения сложных научных задач. В её основе лежит открытая GLM-5.3, которую дополнительно обучили на научных данных, лабораторных записях и материалах исследовательских проектов. Результаты тестирования: • 38,3% в FrontierScience-Research против 33,3% у GPT-6 Astra. • 93% в GPQA Diamond — тесте на сложные научные вопросы. Главная особенность Hydrogen-1.0 — открытые веса. Исследователи смогут изучать модель, адаптировать её под собственные задачи и развивать дальше. [Источник]()
From the original source:
Telegram — GPT Time | Нейросети | ИИ
Read full article at Telegram — GPT Time | Нейросети | ИИAn open neural network for scientific research that beat GPT-6 Astra in the test
🧪 An open neural network for scientific research that beat GPT-6 Astra in the test The developers presented Hydrogen-1.0 - an AI model specially trained to solve complex scientific problems. It is based on the open source GLM-5.3, which was additionally trained on scientific data...
AI Summary & Key Takeaways
- 🧪 An open neural network for scientific research that beat GPT-6 Astra in the test The developers presented Hydrogen-1.0 - an AI model specially trained to solve complex scientific problems
- It is based on the open source GLM-5.3, which was additionally trained on scientific data...
An open neural network for scientific research that beat GPT-6 Astra in the test
Related stories
OpenAI hires ex-Army Secretary, gaining bipartisan reach just as midterms loom
OpenAI hires Dan Driscoll, the former civilian head of the US Army, as "Senior Counselor for National Security Modernization," likely to win government contracts and help shape AI rules for the military and intelligence agencies. Driscoll drove AI projects inside the Army, is a…
OpenAI is quietly renaming data centers ‘compute campuses’
The Stargate AI data center ’compute campus’ in Abilene, Texas. | Bloomberg via Getty Images People hate data centers. They're loud, resource-hungry eyesores that are so profoundly unpopular they have managed to achieve the near-impossible in today's polarized political…
DeepSeek is preparing a blow to NVIDIA: they want to train the new V5 on Huawei chips
🐳 DeepSeek is preparing a blow to NVIDIA: they want to train the new V5 on Huawei chips According to Chinese media, DeepSeek plans to use Huawei Ascend accelerators to train the next V5 model, reducing dependence on American GPUs. Previously, Huawei chips were mainly used...
AI news digest for November 3, 2026
The loudest thing of the day came in a dry legal document: Anthropic warned that it was close to a model with recursive self-improvement. At the same time, the White House took up the dictionary: Trump ordered
