At a time when general-purpose large models are emerging like a flood, artificial intelligence is accelerating into the deep waters of hard technology. On January 26, 2026, Shanghai Jiao Tong University officially launched Optics GPT, a vertical large model in the optical field. This achievement not only marks an important breakthrough in the integration of "AI + hard technology" in China, but also provides a smart "digital brain" for optical research through precise professional insights.

"Senior Expert": Cultivating AI Experts Like Training Doctoral Students"

If ChatGPT is an all-knowing "generalist", then Optics GPT is a "senior expert" deeply rooted in the laboratory. By deeply learning the professional knowledge of the optical field, it can understand optical principles like a "virtual optical expert", providing highly reliable intelligent support for scientific research, design, and teaching.

image.png

To verify its practical capabilities, the team built a specialized evaluation set covering the following six core areas:

Basic theory: optical physics, optical quantum;

Frontier technology: nonlinear optics, optical computing;

Engineering applications: optical design, optical communication.

The evaluation results show that Optics GPT outperforms mainstream general-purpose large models in all core dimensions, fully verifying the technical path of "small-scale specialized models" surpassing "large-scale general models" in vertical fields.

Four Core Features, Redefining Industry Access Barriers

As a model designed specifically for the industry, Optics GPT has the following distinct characteristics:

Lightweight deployment: with a model scale of only 8B parameters, it supports efficient operation on the terminal and edge, greatly reducing the application barriers for small and medium-sized enterprises.

High cognitive level: systematically injects professional knowledge of optics, forming accurate physical intuition.

Strong application implementation: performs leading in core scenarios such as algorithm generation, system diagnostics, and simulation design.

Safe and controllable: the entire process from data construction, model training to deployment is independently controllable, ensuring the data privacy of core industries.

With the release of Optics GPT, the R&D paradigm in the optical field is undergoing a revolutionary change. From tedious experimental assistance to complex system diagnostics, AI is moving from the background to the forefront, becoming the most compatible digital partner for scientists and engineers.