Intern-S1-Pro is the first trillion-parameter scientific multimodal foundation model, outperforming proprietary models on specialized scientific reasoning.
March 27, 2026
Original Paper
Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale
arXiv · 2603.25040
The Takeaway
It demonstrates how to scale RL-based scientific training to the trillion-parameter level across 100+ specialized tasks like chemistry and materials science. It signals the arrival of open-source models that can compete with top-tier proprietary systems in deep domain expertise.
From the abstract
We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, incl