Kai Chen
Leading Young Scientist at the Shanghai AI Lab and Head of the Large Model Center
Kai Chen is a Leading Young Scientist at the Shanghai Artificial Intelligence Laboratory and Head of the Large Model Center. He has published over 70 papers in top-tier AI conferences and journals, with more than 34,000 citations on Google Scholar. He has been recognized with honors such as the Shanghai Oriental Talent Program (Leading Talent) and Shanghai Young Academic Leader. He leads the development of the InternLM large model, taking on key technical challenges, and has established the “Sinan” evaluation system, which is widely adopted by leading large model companies and research institutions. Previously, he served as the head of the computer vision open-source algorithm system at OpenMMLab, where he made significant contributions to the open-source ecosystem and built strong international influence.
Topic
Trillion-Parameter Scientific Multimodal Foundation Model: A Technical Deep Dive into Intern-S1-Pro
**Intern-S1: A Scientific Multimodal Foundation Model Pushing the Frontier of Open Source AI** The Shanghai Artificial Intelligence Laboratory has released and open-sourced the Intern-S1 and Intern-S1-Pro scientific multimodal foundation models. Their general capabilities rank among the top tier of open-source models, while their scientific capabilities reach a globally leading level. These models integrate extensive multidisciplinary expertise, with a strong focus on enhancing scientific reasoning and domain-specific performance. On benchmark tasks across chemistry, materials science, and life sciences, Intern-S1 surpasses leading closed-source models such as GPT-5.2 and Gemini-3-Pro. In addition, Intern-S1 pioneers a new paradigm that unifies general-purpose and specialized capabilities through multi-task co-training. It supports large-scale multi-task reinforcement learning, enabling both broad competency and deep domain expertise simultaneously. **Outline:** 1. Overview of the InternLM model ecosystem and its key components 2. Introduction to the scientific multimodal models Intern-S1 and Intern-S1-Pro 3. Architecture design of the Intern-S1 series and scientific data generation pipelines 4. Training framework supporting trillion-parameter-scale models 5. Key algorithmic innovations in the Intern-S1 series
