--- title: "How one English AI tutor reached 15,000 Chinese schools — and 5.3B cumulative calls" date: 2026-11-07 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-app-tianxue-ai-english-tutor language: en --- # How one English AI tutor reached 15,000 Chinese schools — and 5.3B cumulative calls > Tianxuewang (天学网), a Beijing firm with 23 years in English teaching, has put a subject-specific large model (大模型) into more than 15,000 primary and secondary schools, crossing 5.3 billion cumulative model calls by late October 2026. A teacher in a county high school finishes grading a stack of English essays — not because she rushed, but because an AI read every one first and marked the obvious errors, leaving her to focus on voice, argument, and the handful of students who needed her most. That scene is the kind of use case the company highlights. The system behind it comes from Tianxuewang (天学网), a Beijing Zhongguancun-based national high-tech enterprise that has spent 23 years building English-teaching software. Its bet is narrow and deliberate: instead of a general chatbot, build an English-specific large model (大模型) tuned to the actual rhythm of Chinese school classrooms. ## Key takeaways - Tianxuewang (天学网) says its English AI now serves **more than 15,000 primary and secondary schools** nationwide. - Cumulative large-model calls crossed **5.3 billion by late October 2026** (up from 4.7 billion in June), with the spoken-English assessment engine accounting for more than **4.4 billion of those** calls. - The product is a full teaching loop — pre-class prep, in-class teaching, after-class practice, periodic testing, and precise evaluation — not a single Q&A tool. - It runs inside a national policy window: the "AI + Education" Action Plan (《人工智能+教育》行动计划) issued by the Ministry of Education and four other ministries in April 2026, which targets a systemic change in teaching models by 2030. ## Why a subject-specific model, not a general one General large models (大模型) are impressive at open conversation but weak at the grinding, repetitive work of a subject like English: scoring pronunciation, catching tense errors across a class of forty, pushing the right drilling exercise to the right student. Tianxuewang's argument is that a model trained on the specific structure of English learning — phonetics, writing rubrics, error patterns common to Chinese learners — does the mundane layer better, which is exactly the layer that eats a teacher's evenings. The numbers suggest adoption has been broad rather than boutique. China Daily reports the platform served over 15,000 schools, and that cumulative model calls reached 5.3 billion by the end of October 2026, with the oral-assessment engine accounting for more than 4.4 billion of those. A second, company-profile source puts the school count above 15,000 for 2025 as well, so the scale is not a one-quarter spike. The trajectory — 4.3 billion calls by December 2025, 4.7 billion by June 2026, 5.3 billion by October 2026 — points to steady, school-year-driven usage rather than a launch spike. ## What it actually does in a classroom The system is built around the full instructional loop. Before class, it helps teachers prepare; during class, it supports delivery; after class, it drills and tests; then it evaluates. The payoff is most visible in writing. Chengdu No. 7 High School (成都七中) deployed Tianxuewang's writing system to build a precise-teaching loop, and the case entered the National Smart Education Platform (国家智慧教育平台). Beijing No. 80 High School (北京第八十中学) adopted the standardized speaking-and-listening assessment system and used the automatically generated learning-analytics reports to run targeted, tiered tutoring on each student's weak points. Crucially, the design keeps the teacher in the center. The AI absorbs the mechanical load — first-pass scoring, error flagging, routine practice — while the teacher handles the parts a model cannot: tone, argument, encouragement, and the human read on a struggling student. The framing matters because it directly addresses the fear that AI will push teachers aside; here the selling point is teacher time, not teacher replacement. ## Sitting inside the regulatory rails None of this scales in Chinese schools without compliance, and Tianxuewang has been careful on that front. Its core algorithms are filed with the Cyberspace Administration of China (国家网信办备案), its products passed Ministry of Education qualification review and entered the digital-campus solution library of the Central Audio-Visual Education Center (中央电教馆). In October 2026 the company's full-scene English teaching solution joined the Beijing primary-and-secondary smart-education platform's AI app store, and its essay-grading project won first prize in the digital-education track of the 2026 Third China Internet Innovation Competition. Tianxuewang also became a unit member of the artificial-intelligence subcommittee under the national information-technology standardization body, a signal that vendors are being pulled into standard-setting rather than left to improvise. ## What readers can do now - **Parents and teachers evaluating AI tutoring tools should check the compliance paper trail first** — algorithm filing with the CAC, education-ministry qualification, and inclusion in an official platform are stronger signals than a flashy demo. - **Ask what loop the tool closes.** A model that only answers questions saves little; one that handles prep, practice, testing, and evaluation changes a teacher's week. - **Treat AI as a load-balance, not a replacement.** The useful question is not "can it teach?" but "which repetitive layer does it remove so a human can do the human part?" ## Honest limitations The adoption and usage figures (15,000+ schools; 5.3 billion cumulative calls; 4.4 billion oral-assessment calls) come from China Daily reporting in July and October 2026, with the school count also echoed in a company profile. These are company-stated numbers carried by a state-linked outlet, not independently audited enrollment or usage data, so treat the scale as indicative. I have not verified per-school outcomes or any measured score improvement, and the specific case studies (Chengdu No. 7, Beijing No. 80) are presented by the company and local media rather than in a controlled study. The "AI + Education" Action Plan's April 2026 issuance by the Ministry of Education and four other ministries and its 2030 target are reported by China News Service; I did not retrieve the original ministry text. I have focused on English subject tutoring and have not covered math, science, or vocational AI-education deployments. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-tianxue-ai-english-tutor Free to quote with attribution and a link to the original.