Intern-S2-397B
Shanghai AI Laboratory · 2026-09-13 · 403.4B parameters
Shanghai AI Laboratory's released scientific multimodal foundation checkpoint, distinct from Intern-S2-Preview-397B. It combines visual pretraining with scientific and agent reinforcement learning, accepts text, images and video, and also supports specialized scientific sequences and time-series tasks. The complete checkpoint has 403,423,094,768 parameters and a 262,144-position text configuration; thinking is enabled by default and can be disabled. The weights and model card were published on Hugging Face on September 13, 2026, the same day the official GitHub README added the release.
Benchmark scores
| Benchmark | Score |
|---|
No benchmark scores recorded.