Skip to main content

Shenzhen AI Model: One Brain to Rule DNA, Weather, and Everything Science

A Super Brain for Science

On July 19, at the Scientific Intelligence Open Forum of the 2026 World Artificial Intelligence Conference in Shanghai, the Shanghai Institute for Scientific Intelligence pulled back the curtain on a new kind of AI—one that doesn't just chat or generate images, but tries to make sense of the very fabric of science. Its name, Shenzhen, comes from the "Divine Pearl Iron" in Journey to the West, a fitting metaphor for a tool meant to handle diverse scientific tasks in a compact, powerful form.

Think of it as the brain behind the "Great Sage" system-level intelligent research agent launched earlier this year. Now, that brain is out in the open.

What Makes Shenzhen Special?

Technically, Shenzhen is a multimodal foundation model with about 11 billion parameters. But the real magic is what it can handle: DNA, RNA, proteins, small molecules, Earth systems, and medical images—all within a single model. It's like having a scientist who can switch from genetics to meteorology without missing a beat.

How does it pull this off? The shared backbone uses Qwen3-VL-8B, but each of the six data types gets its own dedicated processing path. Before feeding into the common model, the system preserves the sequence order of biological sequences, the connection structures of molecules, the spatial distribution of weather fields, and the local details of medical images. This means Shenzhen can directly generate RNA sequences, molecular representations (SMILES), global weather maps, and even medical image segmentations—all from one framework.

Open Science, Open Code

One of the most exciting parts? It's open-source. The Shanghai Institute has released model weights, inference code, example scripts, and documentation. Researchers can download the model or call APIs through the Xinghe Qizhi Scientific Intelligence Open Platform, which already hosts over 1,500 scientific models and tools. You can also grab the weights on Hugging Face and the code on GitHub. The message is clear: this is a community effort.

A Stage Full of Big Ideas

The unveiling happened at the Scientific Intelligence Open Forum, co-hosted by Fudan University and the Shanghai Institute. The forum gathered scientists from around the world to discuss how AI might reshape the entire process of scientific discovery.

Nobel laureate Professor Arie Y. K. Wachter stressed that AI must be grounded in reliable physical mechanisms. Turing Award winner Professor Gilles Brassard shared his excitement about quantum computing's potential in new material discovery, but also offered a piece of advice for young researchers: follow your curiosity, not the hype. "Current research findings might only truly take effect after a decade," he reminded the audience.

Wang Jian, director of Zhejiang Lab and an academician of the Chinese Academy of Engineering, pointed out a fundamental shift: scientific intelligence isn't about text—it's about data. "Today's foundation models are still based on text," he said. "In the future, scientific data should become the native inhabitants of scientific intelligence." He envisions a world where data from every discipline is tokenized and brought into a common representation space, making AI as fundamental as mathematics.

Professor Jianqing Fan from Princeton University spoke about AI's role in society, describing it as a dynamic cycle of statistical learning and optimization. He highlighted how AI transforms data into social insights—from measuring economies to managing financial risk—but warned that we must keep an eye on employment shifts, education changes, and ethical concerns.

The Bigger Picture

During a summit dialogue, participants agreed on one thing: the key to original innovation in the AI-native era is creating a new collaborative discovery mechanism that connects people, data, models, experiments, and academic judgment. That means not just upgrading research infrastructure, but also changing how researchers think.

Qi Yuan, a specially appointed professor at Fudan University and director of the Shanghai Institute, laid out the ultimate vision: "AI should move from predicting the next token to discovering unknown laws. The discovery of unknown scientific laws by AI will mark the beginning of super intelligence." He emphasized that real research is a long chain—literature, hypotheses, data, models, simulations, experiments, feedback—and scientific intelligence is about building the systemic capability to organize that entire process.

What This Means

When Shenzhen, with its 11 billion parameters, brings six types of scientific data into one mind, the form of scientific discovery may be quietly rewritten. It's not just a model; it's a glimpse into a future where AI doesn't just assist scientists—it becomes a partner in uncovering the universe's deepest secrets.

Key Points

  • Shenzhen is an 11-billion-parameter multimodal foundation model for scientific data.
  • It handles six data types: DNA, RNA, proteins, small molecules, Earth systems, and medical images.
  • The model is open-source, with weights, code, and APIs freely available.
  • Experts at the forum emphasized AI's role in discovering unknown scientific laws, not just predicting tokens.
  • The vision: AI as a fundamental tool for scientific discovery, integrated with data and human judgment.