Daily AI Digest

2026-09-14

Source:橘鸦 AI 早报 · 2 items

2026-09-14
2026-09-16 2026-09-15 2026-09-14 2026-09-13 2026-09-12 2026-09-11 2026-09-10 2026-09-09 2026-09-08 2026-09-07 2026-09-06 2026-09-05

智谱敲定约50亿美元融资,开发下一代GLM基础模型

要闻

智谱敲定约50亿美元融资,开发下一代GLM基础模型

Zhipu has finalized an approximately US$5 billion equity-and-debt financing plan, comprising about US$2 billion in new H-share placements and roughly US$3 billion in zero-coupon convertible bonds. Around 60% of the net proceeds will fund its next-generation GLM foundation model, fully self-trained system, and compute infrastructure.

On September 13, 2026, Zhipu disclosed an approximately US$5 billion financing plan comprising about US$2 billion in new H-share placements and roughly US$3 billion in convertible bonds. The company plans to place up to 21.965 million new H shares at HK$714.00 per share. The zero-coupon convertible bonds will be issued at 100.5% of their principal amount, with an initial conversion price of HK$892.50 per share, 25% above the placement price. According to the company, approximately 60% of the net proceeds will fund research and development of its next-generation GLM foundation model and fully self-trained system, as well as the deployment and upgrading of large-scale training, production inference, compute resources, and related technical infrastructure. About 15% will be used for business expansion, strategic investments, and potential M&A, while approximately 25% will go toward optimizing its capital structure, replenishing working capital, and other general corporate purposes. The fully self-trained system will explore a recursive self-improvement loop (RSI), covering training-data generation and filtering, the construction of executable and verifiable task environments, long-horizon reasoning, and self-verification. Zhipu will also optimize compute infrastructure through chip adaptation, operator development, and compute scheduling.

Read original →

书生-S2 正式版 Intern-S2-397B 开源

模型发布

书生-S2 正式版 Intern-S2-397B 开源

Shanghai AI Laboratory has released the official Shusheng-S2 model, Intern-S2-397B, as open source on Hugging Face. According to the lab, the model was jointly trained on scientific reinforcement-learning tasks spanning more than 20 fields and has improved its handling of long-horizon scientific tasks.

Shanghai AI Laboratory has listed the official Shusheng-S2 model, Intern-S2-397B, on Hugging Face as open source. Deployment supports multiple inference frameworks, and users can also register to call the official Intern API directly. Compared with Preview-397B, the lab says the official release further optimizes scientific capabilities and handles long-horizon scientific tasks more effectively; it also describes the model as its strongest multimodal foundation model to date for scientific intelligence and long-horizon Agent capabilities. According to its explanation, the model uses visual pretraining to learn directly from original pages of scientific literature, jointly trains on scientific reinforcement-learning tasks across more than 20 fields, and connects to a large-scale sandbox environment for black-box Agentic reinforcement learning. The lab also claims leading general reasoning performance among open-source models and strong results on specialized scientific tasks including biomolecular interaction design and material structure generation.

Read original →