Feixiang Teacher 3.0 lets educators generate AI-powered learning apps from a single prompt

Feixiang Teacher, a China-based education large model product, has released version 3.0, adding the ability to generate fully functional education applications from a single sentence of description.

Users describe what they need and within minutes get an application that can be opened through a link, calls AI models while running, and can be iterated over time. Students click the link and start using it; teachers need no coding knowledge and handle no deployment. Each generated app has its own URL and can support user registration and data persistence.

The company positions the release as a shift from generating content to generating software. Where earlier versions produced teaching animations, courseware and instructional resources, 3.0 puts application development — previously the work of an engineering team — into a chat box.

Two examples illustrate the range. A teacher wanting a daily vocabulary challenge app received "English Adventure Island," complete with a course map, learning levels, listening and speaking exercises, points and rewards, streak tracking and individual progress. Students resume where they left off the next day, rather than finishing a one-off slideshow.

In a second case, a high school student who felt the periodic table should be explored rather than memorized handed his own notes and method to the tool, and got an application that lets users search and filter elements, inspect properties, and compare across categories and periods — turning a static chart into an explorable chemistry tool.

Generated apps follow an education framework and can draw on authoritative question banks and resource libraries. They can also invoke Feixiang's education model capabilities — homework grading, error diagnosis and personalized question generation — so an app performs real teaching tasks rather than only displaying content.

Underlying model upgrades accompany the feature. High-fidelity document parsing better preserves original questions, formulas, illustrations and teaching semantics in textbooks, exam papers and lesson plans, converting uploaded files into editable, explainable courseware or apps. 3D interaction and professional diagram generation can turn a flat cell structure diagram into a rotatable model with clickable labels, or render a complex physics process as a clearer animation. Audio and video upload, comprehension and embedding let teachers bring their own recorded explanations and lab footage into generated materials.

The release also improves access to existing school assets: courseware, materials and data from previously deployed systems can be pulled into new generation workflows, rather than sitting scattered across files and platforms. Question setting, grading, error diagnosis and subject explanation were further trained and optimized, with the company arguing that education capability runs through model training, capability invocation and application runtime instead of being bolted on at the end. Layered memory, a personal resource library, one-click repair and version history round out the update.

"From teaching animations and interactive courseware to generating education applications, what we want to do hasn't changed," the product lead said — keep expanding what education AI can do and build the best large model product for education scenarios. The company argues the creator of education apps is expanding from professional engineering teams to anyone with genuine domain understanding: teachers, parents, students, researchers and institutions alike.