InternLM (Shusheng Series)

InternLM (Shusheng Series) is a family of open-source models continuously released by Shanghai Artificial Intelligence Laboratory. It currently includes the language model InternLM3, the scientific multimodal model Intern-S1, and the strong reasoning model InternThinker. The series follows a “general-specialist fusion” approach to improve data efficiency and domain performance, providing a reproducible AI foundation for research and industry.

In terms of capabilities, InternLM3 achieves performance comparable to models trained on 18T tokens while using only 4T tokens, reducing training costs by 75%; it also integrates conventional chat and deep thinking modes in a single model. Intern-S1, the first open-source general-purpose scientific multimodal model, excels in interdisciplinary analysis. InternThinker demonstrates long chain-of-thought and self-reflection, showing strong results in math, coding, and complex reasoning tasks. All models are open-sourced on GitHub, HuggingFace, and ModelScope, and can be used via local deployment or API calls.