InternLM Releases Science-Specialized AI Model Intern-S2-397B

The InternLM team has released Intern-S2-397B, their most powerful multimodal foundation model designed for scientific intelligence and long-horizon agent tasks. The model has been scaled across three key dimensions: pre-training, reinforcement learning task coverage, and interactive agent environments.

Key Technical Features

• New Pre-training Paradigm: Visual pre-training directly from original pages of scientific literature jointly models symbolic semantics and visual relationships in a shared representation space without intermediate parsing. This preserves connectivity between text and visual elements and improves data efficiency.
• Scientific Modality Reasoning and Generation: Large-scale multi-task reinforcement learning across more than 20 diverse scientific domains has achieved top-tier general reasoning performance among open-source models. It shows particularly strong results in specialized scientific tasks such as biomolecular interaction design and material structure generation.
• Long-Horizon Agents: By connecting multiple agent frameworks to large-scale sandbox environments for black-box agent reinforcement learning, the model improves generalization capability and performance ceilings for long-duration tasks in both general and scientific domains.