Around the world · Updated October 2026
What other schools are trying, and what happened
Alpha School in Texas and China’s national push get the most attention. We looked at both, plus South Korea, Estonia, Singapore and others. For each: what it does, what is known, what is only claimed, and what a teacher with no special budget can borrow.
How to read this page
Marketing is not evidence.
We mark each claim: Claimed means the program says it. Reported means a news outlet or official source reports it. Documented means published research or an official document. Many figures here come from media reports and company statements, and several rollouts are too new for independent results.
Explore
Pick a program
What it is
Two hours of core academics a day
Alpha School is a private school network that began in Austin, Texas. Its “2 Hour Learning” model has students work on adaptive software for about two hours of core academics each morning, with adults serving as “guides” who motivate and support rather than lecture. Afternoons are for workshops, projects, arts and sports. Alpha calls its software platform “TimeBack.”
Claimed by Alpha.
The results claim
“Top 1%” and “2x learning”
Alpha says its students “learn 2x” and rank in the top 1% nationally; reviewers also quote an Alpha figure of 2.6 times normal growth on MAP tests. These results are self-reported, without a control group or an independent audit, and Alpha has not released the underlying data. Students may already have been high achievers when they enrolled. A fair reading is: unproven, not disproven.
Claimed No independent study found
How a morning works
Small lessons, instant feedback, a gate before moving on
- Daily minimums. Each day has a set of required lessons. Progress is counted in “XP,” where 1 XP is meant to equal about one minute of focused work; the target is about 120 XP a day. Reported
- Feedback on every answer, with an explanation when it is wrong. Reported
- A mastery gate. Earlier reporting described lesson credit at 80% correct; Alpha now says students must show over 90% accuracy before moving on. Reported Claimed
- Missed items come back later, just before you would forget them (spaced repetition). Reported
- Coaching calls. When a student keeps stumbling, the system asks them to book a call with a remote teacher. Reported
- Visible progress, including an estimate of when a student will finish each subject’s grade level. Reported
- Some apps start with a placement test: Math Academy, for example, begins with an adaptive diagnostic exam. Claimed by Math Academy.
From a parent’s review on Astral Codex Ten (2025), Austin Scholar (2025), Wired (2025) and Alpha’s website.
The app stack
Mostly practice apps, not chatbots
| Subject | Apps reported in 2025 |
|---|---|
| Math | Synthesis and AlphaMath Fluency (K–3); Math Academy (grades 4–12) |
| Reading | AlphaRead, then TeachTales, then long-form books (grades 3–8) |
| Writing | AlphaWrite (grades 3–8) |
| Vocabulary | Membean (grades 6–12) |
| Science | MobyMax (grades 3–5); Khan Academy (grades 6–8) |
Earlier, a single commercial practice app, IXL, carried most of the academic work. One newsletter reports that Alpha’s own apps used fixed, human-checked content through 2025 and began generating content with AI in 2026.
Reported Austin Scholar (August 2025), Astral Codex Ten review.
The part that isn’t AI
Motivation does a lot of the work
- At one Alpha-model campus, students earn a school currency only when they finish their lessons with 80% or more.
- “Check charts” list tasks a student must complete, often in front of others, before moving up a level.
- Afternoon workshops aim at real, visible outputs, though some parents at the flagship campus called them not academic enough.
- The parent review reports that a homeschool pilot using the same software, without guides or incentives, advanced at about the normal rate.
- A teacher-blogger who averaged Alpha’s newer figures found growth about the same as before its switch to generative AI (about 2.5x versus 2.6x), which points to the setup, not the AI.
Reported about Claimed figures. Astral Codex Ten review (2025); Dylan Kane (2026).
The concerns
What reporters found
- Wired (October 27, 2025): IXL said Alpha’s account was deactivated “this past July” for violating its terms of service. Of 31 “coaches” on a 2023–24 list, 26 had LinkedIn profiles showing work for Trilogy or Crossover, companies linked to Alpha’s co-founder, and at least 27 lived outside the US. Parents who want to stop recordings of their children outside school must opt out of an “anywhere” setting.
- 404 Media (February 17, 2026): former employees and internal company documents described AI-generated lesson plans that were faulty and sometimes did “more harm than good.”
- Scientific American (August 28, 2026): Alpha “has not released the underlying data behind its claims.”
- ProPublica and The Texas Tribune (September 2, 2026): Texas’s State Board of Education rejected Alpha’s charter pitch in 2025, one of at least nine states to turn it down; Texas later approved it for its voucher program. In Arizona, the one approved virtual charter (Unbound Academy) reported first-year proficiency of 28% in English and 10% in math.
- Tuition is reported to range widely, with about $40,000 typical at many campuses.
Reported
Worth knowing
It is mostly not “AI chat”
Much of what students use is adaptive practice software of the type many schools already know, not chatbots writing for them. “AI-powered” can mean very different things, so always ask: what does the software actually do, and where is the evidence?
What ARQIV borrowed
The well-tested parts, nothing else
- Your AI Path: short mastery checks with “Skip for now,” and missed questions that come back for review.
- An effort coach in our labs that notices rushing from timing alone, never by camera, and never saves it.
- The Challenge Chart: proof in front of real people, with no currency or ranking.
- Left behind: webcams and tracking, unreviewed AI content, cash-style rewards, pressure metrics and unaudited claims.
Our take. Several ideas behind the model are well supported on their own: mastery learning, immediate feedback, spaced review, clear daily goals and a protected block for real-world skills. Paying for effort students control has research behind it too Documented. The surveillance, the unreviewed AI content, the unaudited claims and the cost are not ideas to copy.
What changed
A national push, in stages
- May 2025: China’s Ministry of Education issued guidelines for AI education in primary and secondary schools and a separate guide for students’ use of generative AI.
- September 2025: Beijing required at least eight class hours of AI instruction per school year. (The eight-hour figure is Beijing’s, not a national rule.)
- April 2026: the Ministry of Education and four other departments issued an “AI+ Education” action plan that aims for an AI-literacy system from kindergarten to university by 2030.
Documented
The age tiers
Perceive, apply, create
Primary: perceive and understand AI. Primary pupils are not to use open-ended content generators on their own, and use AI only with teacher supervision.
Middle school: apply AI for tasks such as analysis and exploration, in stages.
High school: create and innovate, including inquiry learning and technical research. Students are told not to have AI do their homework.
Reported Check the official translation for exact wording.
Exam time
Turning tools off for the gaokao
During the national college entrance exam in June 2025, several major Chinese AI apps reportedly disabled photo-recognition or chat features during exam hours. A tool’s makers can switch it off when integrity matters. A school cannot do that with an app on a phone.
Reported
The open questions
No independent results yet
- We found no independent evidence of learning outcomes from the 2025 push, which is still new.
- Reports describe a booming market for AI learning tablets, with concerns about pressure on parents to buy them, student data and company-reported results.
- Practical limits include teacher training, large rural classes and China’s own limits on screen time.
What to borrow. The tiered idea is useful: young children learn what AI is and how it works before they use it, middle schoolers use it in structured ways, and older students create with it. That is close to how our curriculum guide is built.
| Place | What happened | Lesson |
|---|---|---|
| South Korea | AI digital textbooks launched in March 2025. In August 2025 the National Assembly removed their status as official textbooks, making them optional supplementary materials. Reported problems included errors, privacy concerns and teacher workload. | A fast national rollout without piloting and teacher buy-in struggled. |
| Estonia | “AI Leap” (TI-Hüpe) began in September 2025 for upper secondary students and teachers, with a learning tool designed to guide thinking rather than hand out answers. Early usage data from spring 2026 showed steady but not universal use. No independent learning results yet. | Train teachers first, design the tool not to give answers, and roll out in stages. |
| Singapore | In February 2026 the education minister said the ministry is formally studying over-reliance and holding some AI tools back from lower primary pupils. | Protect foundations before adding tools. |
| Finland | National recommendations on AI in education from early childhood through adult education, published March 2025. | Guidance can be national while choices stay local. |
| United Arab Emirates | An AI curriculum for public schools from kindergarten through Grade 12 began in 2025–26. | Whole-system coverage, results still to come. |
| India | CBSE launched a computational thinking and AI curriculum for classes 3–8 on April 1, 2026, using activities and puzzles. | Teach thinking skills first, then tools. |
| England, Ireland, Australia, Japan | National guidance documents for schools: England’s Department for Education guidance (first published March 2023 and updated since, including January and August 2025; Scotland, Wales and Northern Ireland set their own), Ireland (October 2025), Australia’s framework (2023) and Japan’s guidelines (December 2024 update). | Most stress teacher judgment, privacy and teaching about bias and hallucination. |
| Miami-Dade, USA | The district blocked chatbots in 2023, then gave Google’s Gemini to more than 100,000 high school students after training for educators. No independent outcome data found. | Note the vendor partnership. Ask for evidence. |
Khan Academy’s Khanmigo
Modest gains, light use
A two-year randomized trial in Tennessee middle schools reportedly found a small math gain similar to Khan Academy practice without AI, and students rarely used the chat tutor. Khan Academy’s own Newark claims are the company’s. Common Sense Media rated AI teacher assistants as “moderate risk.”
Working paper Read the paper before quoting numbers.
Does AI cause cheating?
Not much more than before
Challenge Success surveyed about 30,000 high-school students. Roughly the same share (about a third) copied a classmate’s homework as used AI as an unauthorized aid in the past month. The takeaway was to redesign assignments and build motivation, not to rely on bans or detectors.
Reported
From Alpha
Mastery and short blocks
Do not move on until a skill is solid. Use 25 to 30 minute focused blocks, daily visible goals and quick feedback from a paper answer key or a peer check. No software needed.
From Alpha
A real-world block
Protect time for public speaking, money, teamwork and projects. Life-skills time is where students use what they learn.
From China
Age tiers
Understand before use, use under supervision before independence, create last.
From Estonia
A tutor that asks, not tells
Use tutor prompts that give hints, not answers. Train teachers first. Get the prompts →
From South Korea
Pilot first
Try one class for six weeks. Measure learning without AI. Then decide.
From Singapore
Protect the foundations
Handwriting, number sense and reading stay screen-free for young children.
Do not copy: student surveillance (webcams, biometrics), unaudited success claims, rushed system-wide rollouts, and any tool you cannot show the teacher the data practices of.
Sources
Where this comes from
These were compiled from news reports and official documents in October 2026. Where we found an official document we link it; entries marked “search the headline” still lack a verified direct link, so treat those details as reported, not confirmed. Open the sources before quoting a number. Some outlets limit automated access, so we could not read every article in full.
- Wired, Todd Feathers, “Parents Fell in Love With Alpha School’s Promise. Then They Wanted Out,” October 27, 2025.
- Astral Codex Ten, “Your Review: Alpha School” (a reader’s review, June 27, 2025).
- Austin Scholar #177, “The Alpha App Stack,” August 10, 2025.
- Dylan Kane, “Alpha School’s Secret Sauce,” February 19, 2026.
- Emanuel Maiberg, 404 Media, “‘Students Are Being Treated Like Guinea Pigs:’ Inside an AI-Powered Private School,” February 17, 2026 (paywalled; we read the opening).
- Mary Randolph, Scientific American, “Alpha School’s AI teaching model is expanding. Does it work?”, August 28, 2026.
- Lexi Churchill, ProPublica and The Texas Tribune, “Several States Rejected This Private School Chain as a Charter. Now It’s Eligible for Texas’ Taxpayer-Funded Vouchers,” September 2, 2026.
- Math Academy, “How Our AI Works.”
- Alpha School website (source of Alpha’s own claims).
- Roland G. Fryer Jr., “Financial Incentives and Student Achievement: Evidence from Randomized Trials,” NBER Working Paper 15898 (2010).
- Ministry of Education (China), guide to generative AI use in primary and secondary schools, May 2025, in a Georgetown CSET translation.
- Xinhua (english.gov.cn), Beijing to introduce AI courses across primary, secondary schools: at least eight class hours of AI instruction per academic year from the fall semester, March 12, 2025.
- Ministry of Education of China and four other departments, “AI+ Education” action plan, April 2026 (ministry release, in Chinese).
- Rest of World, coverage of South Korea’s AI digital textbooks and China’s AI learning devices. (search the headline; no direct link verified)
- TI-Hüpe (AI Leap), Estonia.
- Finnish National Agency for Education, AI recommendations for early childhood education, teaching and training published, 2025 (in Finnish).
- Press Information Bureau, Government of India, release cited for the CBSE computational thinking and AI curriculum.
- Department for Education (England), Generative artificial intelligence in education (first published March 2023, updated since).
- Ministry of Education, Culture, Sports, Science and Technology (Japan), generative AI in education (in Japanese).
- “One Click Away: AI Tutoring with Khanmigo,” EdWorkingPapers (working paper).
- Common Sense Media, AI teacher assistants risk assessment, 2025.
- Challenge Success, Stanford Graduate School of Education, student cheating survey data. (search the headline; no direct link verified)