---
title: "TAL's AI tutor is now in 50+ Chinese schools — and graded by a government lab"
date: 2026-10-27
category: AI in Action
site: NeuroAI
canonical: https://neuroai.site/a/na-ai-tutor-china-schools-rollout
language: en
---

# TAL's AI tutor is now in 50+ Chinese schools — and graded by a government lab

> TAL Education's Jiuzhang AI tutor app runs dual virtual teachers in 50+ schools, and its model earned the top rating in a CAICT trusted-AI test.

A 13-year-old in a Hangzhou classroom points a tablet at a geometry problem. Instead of the answer, the screen asks a question back: "If we move this vertex, what happens to the auxiliary line?" The coach is not a human substitute teacher but "Xiao Si" (小思), one of two virtual teachers inside TAL Education's Jiuzhang Aixue app (九章爱学).

The app is no longer a pilot curiosity. TAL (好未来, formerly Xueersi 学而思) says Jiuzhang Aixue has been rolled into more than 50 schools across China, and its learning tablets pair the company's self-built Jiuzhang large model (九章大模型) with DeepSeek to run a full "do–check–explain–practice" loop at home. For the first time, a Chinese tutoring company's model has also been put through a government-lab capability rating rather than its own marketing.

## Key takeaways

- TAL's Jiuzhang Aixue app runs "virtual dual teachers" (虚拟双师) and is deployed in **50+** schools nationwide.

- The Jiuzhang large model, launched in 2023, was among China's first filed education large models and earned the highest (**4+**) rating in a CAICT (中国信通院) trusted-AI evaluation.

- A Beijing Normal University co-built model, Shicheng Wanxiang (师承万象), is trialed in **60+** Beijing schools via the municipal education commission's AI store.

- The stated goal is the "impossible triangle" — large-scale, high-quality, personalized — not replacing teachers.

## The product that sits at home

The better-known face of this push is the learning tablet. TAL's T4 flagship machine, built on the Jiuzhang model plus DeepSeek, holds roughly 16 million minutes of course content and can walk a student from a wrong answer to a step-by-step explanation and a follow-up drill. The newer T6 series, launched in 2026, added a "Peiyou AI tutor" (培优AI家教) that does step-level attribution — it does not just flag the mistake, it traces *why* the student erred and builds a path forward. TAL says the T6 passed CAICT's trusted-AI smart-tutor evaluation at the 4+ level in June 2026, the same top grade its "Xiao Si AI 1-on-1" (小思AI 1对1) agent already holds.

## Inside the school

The classroom side is where the deployment claim gets interesting. Jiuzhang Aixue's "virtual dual teachers" split the job: one guides reasoning with follow-up questions, the other breaks down the mechanics of a solution. TAL says the system is in use across Beijing, Shanghai, Jiangsu, Zhejiang and Guangdong pilot schools, and the company's own accounts put the national footprint at over 50 schools. Separately, the Beijing Normal University–TAL model Shicheng Wanxiang, built to the new curriculum standards, entered the Beijing education commission's AI application store and is being trialed in more than 60 schools in Haidian and Dongcheng districts.

## What the model is not allowed to do

The vendors are careful, and the caution is partly regulatory. China's after-school-tutoring rules still shape what these products can be, and the design leans on "augment, don't replace." The product manager for Jiuzhang Aixue's teacher edition put it directly: a teacher-review mechanism is built in because education demands higher accuracy than a raw model gives, and "AI does not replace the teacher, it strengthens the teacher's content creation." The 2026 TAL monthly report quotes the same theme from its engineers — the end state is human-led, machine-assisted.

## Why a tutoring company built its own model

TAL's path is unusual: it ran an AI Lab from 2017, was named in 2019 to host a national smart-education AI open-innovation platform, and filed Jiuzhang as an early education large model in 2023. The bet is that generic chatbots solve the wrong problem in education — they answer, they don't diagnose a student's misconception. A subject-specific model wired to two decades of the company's teaching material is the differentiator it is pitching to schools and parents alike.

## Honest limitations

- The **50+** school figure comes from TAL and TAL-backed event coverage; we found no independent audit of learning outcomes tied to those deployments.

- The CAICT **4+** is a *capability* rating of the system's functions, not proof that students learn more. No peer-reviewed efficacy study was located in our search.

- Reach is uneven. These tools assume a tablet, connectivity and a parent who can pay — the classic digital divide, which limits how "large-scale" the personalization really is.

- Tutoring context matters: China's strict rules on academic tutoring still bound what these products may do, so claims of "replacing a tutor" overstate the permitted scope.

## What readers can do now

- Parents: treat the tablet as a reasoning coach, not a crib sheet — use the "ask-back" feature to build understanding rather than collecting answers.

- Teachers: explore municipal AI stores (Beijing's is one example) where tools have at least been vetted before classroom use.

- Watch for independent outcome studies; a 4+ capability score is necessary but not sufficient evidence that the "impossible triangle" has actually been solved.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-ai-tutor-china-schools-rollout
Free to quote with attribution and a link to the original.
