Intern-S2-Preview
35B scientific model continued from Qwen3.5 that matches the 1T Intern-S1-Pro on core science tasks and generates material crystal structures.
Scales the difficulty and coverage of science tasks through pretraining and RL ('task scaling') with MTP and chain-of-thought compression in RL. Apache-2.0.
- Date
- Friday, 15 May 2026
- Lab
- Shanghai AI Laboratory (InternLM)
- Kind
- open-weights
- Access
- open weights
Date from HF weights upload. Comparison to Intern-S1-Pro is the lab's claim.
Sources
This record was checked against its sources on 6 October 2026. How we check