iFlytek upgrades Xingguang 2.0, a teacher super-agent built to run the whole teaching chain

iFlytek has rolled out Xingguang 2.0, an upgrade of its teacher-focused super-agent, with the stated goal of linking lesson preparation, classroom interaction, learning assessment, home-school communication and professional development into a single chain. The release lands around China’s Teachers’ Day and runs on the company’s Spark X2.5 large language model, drawing on more than two decades of iFlytek’s work in smart education.

The positioning reflects a broader shift in agent technology. Where earlier classroom AI answered discrete questions, newer systems are expected to decompose tasks, call tools and close the loop on a workflow. Education raises the bar: curriculum standards, lesson delivery, diagnosis of student progress, parent communication and teacher training are tightly coupled, and each demands subject accuracy, pedagogical soundness and data security.

In lesson preparation, Xingguang 2.0 generates interactive H5 resources from plain-language instructions, so teachers need neither design nor coding skills. The company cites English picture-book reading, dynamic geometry demonstrations, science experiment simulations, structural model breakdowns and quiz interactions as supported formats. iFlytek’s own example prompt asks the agent to build an interactive animation of Alice in Wonderland paired with scenario-based reading questions.

The resources are not endpoints. Once a class interaction finishes, generated, self-made or saved H5 materials can be converted into teaching applications with one click. With data collection switched on, student responses and task completion feed directly into the class-level learning profile, so in-class activity is archived as assessment data rather than disappearing when the bell rings.

To cut the time teachers spend hunting for material across platforms, the product connects to external content ecosystems and aggregates teaching assets from mainstream video and social platforms, then filters, sorts and labels them by source.

On assessment, Xingguang 2.0 builds a process-oriented incentive system. A teacher can leave a one-sentence comment in WeChat — for example noting that a student helped a classmate and awarding five stars — and the system records the evaluation, adds the points and files the entry. Batch comments, quick one-line comments and automatic evaluation inside H5 activities are also supported, alongside a class pet that grows as students accumulate points, turning incremental progress into visible feedback.

The upgrade also introduces Xingguang Jiaxiao, a WeChat mini-program for home-school coordination. Teachers can push homework, notices and information-collection forms in one action, and the system automatically compiles lists of unread, unsubmitted and unacknowledged items. On the parent side, assignments, school notices and check-in receipts are aggregated into one view, removing the need to scroll back through chat history.

For professional growth, iFlytek worked with the team of Professor Wei Rui at Beijing Normal University to build a project-based learning mentor, currently in internal testing. It simulates a research roundtable with a moderator, subject specialists and project-design specialists, guiding teachers to break down complex teaching tasks on a visual whiteboard and reflect on decision points — with the aim of producing a usable course plan while teaching transferable design method. A companion task-based program, Xingguang Academy, is designed to keep building teachers’ AI skills over time.

iFlytek says the point is not to replace teachers but to absorb administrative load while keeping instructional design and project-based thinking under professional guidance. Teachers who log into the Xingguang website or mini-program during the Teachers’ Day period can claim a set of back-to-school benefits.