# INTERN-S1: A Scientific Multimodal Foundation Model

Source: https://www.youtube.com/watch?v=-fhtj8GUGXA
Recap page: https://rapidrecap.app/video/-fhtj8GUGXA
Generated: 2026-02-06T21:37:13.321+00:00

---
## Quick Overview

InternS1, a scientific multimodal foundation model from Shanghai AI Laboratory, significantly outperforms generalist models like GPT-4 and Gemini 2.5 Pro on niche scientific reasoning tasks by employing a novel dynamic tokenization approach, exemplified by its ability to read and structure data from scientific PDFs far better than standard tokenizers.

**Key Points:**
- InternS1 significantly outperforms generalist models like GPT-4 and Gemini 2.5 Pro on specialized scientific reasoning benchmarks.
- The model achieves this through a novel dynamic tokenization method that intelligently processes data structures like tables, diagrams, and equations.
- InternS1 was trained on 241 billion parameters derived from both general text and high-quality scientific data, specifically tailored for scientific tasks.
- The dynamic tokenization compresses information by using a single token for complex structures (like a benzene ring) instead of multiple tokens, drastically improving efficiency.
- The model achieved a 70% improvement in performance compared to GPT-4 on specific benchmarks, such as predicting crystal structure stability.
- The researchers developed a custom algorithm called 'Oriel' to implement this dynamic tokenization, which penalizes unnecessary tokenization collapses.

![Screenshot at 00:23: The speaker highlights a significant failure point in general models: their performance collapses when dealing with the structured data common in niche scientific fields, contrasting this with InternS1's architecture.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-00-23.jpg)

**Context:** The video discusses the InternS1 model developed by the Shanghai AI Laboratory, focusing on its unique architecture designed to handle complex, niche scientific data, which often presents challenges for standard Large Language Models (LLMs) trained primarily on general text. The core innovation lies in its dynamic tokenization strategy, which moves beyond simple word-level tokenization to better understand the structural context found in scientific literature, such as chemical structures and mathematical equations.

## Detailed Analysis

InternS1, a multimodal foundation model from the Shanghai AI Laboratory, demonstrates superior performance over generalist models like GPT-4 and Gemini 2.5 Pro, particularly in complex, niche scientific reasoning tasks. The key differentiator is its architecture, which incorporates dynamic tokenization, allowing it to process structured scientific inputs—such as chemical structures, equations, tables, and diagrams—more effectively than models that rely on standard text-only tokenization. The model was trained on 241 billion parameters, incorporating both general text and high-quality scientific data. The researchers report an improvement of up to 70% on specific benchmarks compared to GPT-4. For instance, while standard tokenizers might break down a chemical structure like a benzene ring into several random tokens, InternS1’s dynamic tokenizer compresses this structure into a single, meaningful token, reducing noise and improving accuracy. The system uses a custom algorithm called 'Oriel' to manage this tokenization, penalizing collapses to a single expert when a generalist approach might suffice, thus ensuring a balance between specialized knowledge and generalization. This approach allows InternS1 to handle tasks like predicting crystal structure stability and interpreting complex molecular synthesis plans with higher fidelity than its predecessors.

### Model Introduction and Performance

- InternS1 outperforms GPT-4 and Gemini 2.5 Pro on niche scientific tasks
- Achieved 70% improvement on certain benchmarks compared to GPT-4
- Outperforms generalist models in handling scientific data structures

### Key Innovation - Dynamic Tokenization

- Uses dynamic tokenization, unlike standard word/character tokenization
- Compresses complex structures (e.g., benzene ring) into single tokens
- Implemented via a custom algorithm called 'Oriel' that manages tokenization scope

### Training and Application

- Trained on 241 billion parameters using a mix of general text and scientific data
- Successfully predicts crystal structure stability
- Excels at interpreting complex chemical synthesis planning and molecular interactions

### Comparison to Base Models

- Base models often fail on scientific data because they treat it as unstructured text, leading to routing collapse or poor accuracy
- InternS1's approach provides a crucial advantage in specialized domains

![Screenshot at 00:00: The initial screen displays the podcast branding for 'ReallyEasy AI' and an invitation to 'Become A Member Today!' against a backdrop of an audio waveform.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-00-00.jpg)
![Screenshot at 00:54: A comparison is made between the generalist approach \(like GPT-4\) and specialized models, setting up the problem of handling complex scientific data.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-00-54.jpg)
![Screenshot at 01:17: The paper details the massive scale of InternS1, noting it has 241 billion parameters.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-01-17.jpg)
![Screenshot at 03:03: The speaker introduces the concept of how InternS1 handles data by using an encoder that compresses sequential signals into a digestible format.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-03-03.jpg)
![Screenshot at 07:46: The speaker highlights the massive scale of the pre-training data used for the model, totaling 2.5 trillion tokens.](https://ss.rapidrecap.app/screens/-fhtj8GUGXA/00-07-46.jpg)