Free for learners / ~$60 per student per year for districts
The category-defining AI tutor. Free GPT-4o access via OpenAI's nonprofit deal makes it impossible for VC-backed startups to compete head-on. Sets the procurement price ceiling.
Track Atlas · OPC ATLAS
AI Math Tutor: After Khanmigo Went Free, Where Is There Still a Business?
Khan Academy gave the headline product away. The wedge moved to verticals, languages, and procurement.
When OpenAI handed Khan Academy free GPT-4o access in 2024, the consumer math-tutor war effectively ended. Khanmigo became a Socratic K-12 tutor that no startup can undercut on price. By 2026 the surviving businesses split into three: (1) photo-OCR incumbents like Photomath (acquired by Google in 2022) that own the snap-to-solve habit on 300M+ phones; (2) premium opinionated tutors like Synthesis ($30/mo, founded by Chrisman Frank ex-Ad Astra) selling structured curriculum to ambitious parents; (3) teacher-side AI like MagicSchool that hit 4M+ teachers across 130 countries by mid-2025. The real white space is not "another GPT-4o wrapper that tutors algebra in English." It is non-English math (Korea, Japan, Latin America), state-specific standards (CCSS/TEKS/EngageNY), and procurement-grade compliance that Khanmigo's flat free tier doesn't serve. Capital required: $300K-3M. Window: 18-24 months before district RFPs lock in incumbents.
01 · 2026 Market Reality
The pricing floor collapsed in 2024 and never came back. Khanmigo costs $0 for any student because Khan Academy gets GPT-4o credits from OpenAI; Google Search itself does step-by-step math now via Gemini; Photomath, which Google bought for an undisclosed nine-figure price in 2022, is bundled into Google Lens and Search. That means no consumer startup can charge for "help me solve this problem." The money has moved up-stack. Three live segments: (a) Procurement — districts paid $13B+ on edtech in the US alone in 2024 (EdSurge data), and Khanmigo for Districts launched at ~$60/student/year, which is the new effective ceiling. Any startup selling to schools must beat that price or differentiate on safety/curriculum alignment. (b) Premium 2C — Synthesis Tutor (~$30/mo) and Brilliant (>$100M ARR, building Mathigon-style interactive curriculum) prove that "structured opinionated curriculum" still commands a price tag above zero. (c) Teacher-side — MagicSchool, Diffit, Curipod sell to teachers as buyers (not students), with seat-based district contracts in the $5-15K range. The exits-to-watch: Magic School raised $14M+ in 2024 and reportedly closed a much larger 2025 round. Synthesis pivoted from cohort school to AI tutor in 2023 and is the cleanest premium 2C case study.
02 · Companies to Watch
300M+ downloads · bundled in Google Lens
The snap-a-photo OCR habit. Founded by Damir Sabol in Croatia, acquired by Google after eight years of indie growth. Distribution moat is the camera-first reflex that pre-GPT users already learned.
$30/mo subscription · founded by Chrisman Frank
Built by the team behind the original Ad Astra/SpaceX school. Premium opinionated 1:1 voice tutor with proprietary curriculum, sold to ambitious parents. The clearest case that paid 2C still works above Khanmigo's free tier.
Industry estimates ~$100M ARR
Not LLM-first — they bought Mathigon in 2023 to bring real interactive courseware in-house. The reminder that "great math UX" beats "chatbot that does math" for users who actually want to learn.
4M+ teachers / 130 countries
Sells to teachers as buyers, not students. Math lesson planning, IEP-aware accommodation generation, worksheet differentiation. Free top-of-funnel; districts unlock at seat-based pricing. Founder Adeel Khan came out of K-12 ops.
100K+ teachers in first year
Differentiation engine — any math passage adapted to multiple reading levels with auto-generated questions. Founder Carolyn Choi is ex-Newsela; the wedge is teacher-time saved on reading-grade differentiation, a problem GPT alone doesn't solve cleanly.
100M+ users · global · Asia-strong
The most-downloaded photo math app outside China. ByteDance distribution + 24/7 live human tutors as upsell. The reminder that "cheaper labor + AI hybrid" still wins in price-sensitive geographies Khanmigo doesn't prioritize.
Step-level reasoning · LMS integrated
Pre-LLM tutor that did step-correct algebra coaching before GPT. Quietly sells into Pearson/McGraw curriculum. A reminder that "deep math correctness" is a real moat against general LLM hallucination.
03 · Green Lights & Red Flags
🟢 Green light · Consider entering
You have a non-English market and a local curriculum standard
Khanmigo's rollout outside English is slow and prioritized by Khan Academy's partner choices. Korean middle-school math, Brazilian ENEM prep, Japanese cram-school workbooks — these have specific curriculum alignment that a localized AI tutor can win on. You don't compete with Khanmigo if Khanmigo isn't even in your language.
You have K-12 procurement experience or a district network
$60/student/year × even 5 mid-sized districts is $1-3M ARR. The buyers are superintendents, math supervisors, and curriculum directors — not students. If you've sold to schools before (ex-Pearson, ex-Renaissance, ex-IXL), the network compounds. If you haven't, the sales cycle is 9-18 months and brutal.
You ship a teacher-side workflow product, not a student chatbot
MagicSchool, Diffit, and Curipod prove the "tools for teachers" layer works because the buyer (teacher or district) and user (teacher) are aligned, and pricing pressure from Khanmigo doesn't apply. Worksheet differentiation, IEP accommodation, parent-comm drafts — boring jobs LLMs do well.
🔴 Red flag · Hold off
You're building a consumer GPT-wrapper math tutor in English
Khanmigo is free. Google Search now does step-by-step. Synthesis owns the premium English-language slot. There is no consumer 2C math tutor wedge left in English; you will spend $100K on paid acquisition and learn this the hard way.
Your differentiation is "ours hallucinates less"
Every LLM lab is shipping reasoning-improved math weekly. By the time you finish your eval harness, the gap will close. Don't bet a company on a moving model-side moat.
You can't articulate a state standard or exam name in your pitch
CCSS, TEKS, EngageNY, GCSE, Gaokao, NCEA, Suneung, JMO, KhMAT — district buyers and tutoring parents both care about specific exams and standards. If your pitch is "an AI tutor that helps with math," you have no procurement story.
04 · Three Ways In
Non-English local-curriculum tutor
Founder with native language + local exam expertise, ideally ex-teacher or ex-edtech
Teacher-side workflow wedge
Engineer + ex-K-12 educator co-founder pair
Premium opinionated 2C tutor
Strong personal brand or media reach in education / parenting
05 · Founders Who Fit Best
Industry Veteran
If you spent 10+ years in K-12 — at Pearson, McGraw, Renaissance, a state board, or as a curriculum director — the procurement network is the moat. Engineering hires are easy; district-superintendent rolodexes are not.
35 and Still Building
Ex-teacher with strong technical literacy is the ideal MagicSchool/Diffit founder profile. Teacher-side wedges reward people who've actually written a math IEP at 11pm.
The Lone Engineer
You can build a math tutor in a weekend. That's the problem — so can 50,000 others. Without a curriculum/teacher co-founder, you have no procurement story and no premium-2C story.
06 · What's Next
Worth reading
- Khanmigo for Districts: pricing & rollout Khan Academy
- 2024-2025 State of AI in K-12 EdSurge
- Synthesis Tutor launch discussion Hacker News
Communities
People to follow
- Sal Khan (Khan Academy) @salkhanacademy
- Adeel Khan (MagicSchool) @adeelk93
- Chrisman Frank (Synthesis) @chrismanfrank
Adjacent tracks
- AI Teacher Tools / Worksheet AISame buyer (teacher/district), broader workflow. Most math-tutor companies eventually expand here.
- Test PrepHigh-stakes exam prep is where parents pay premiums and Khanmigo's flat tier is weakest.
- Paid CommunityAdjacent monetization — math-tutoring brands often layer a paid parent/educator community on top.
Which kind of founder are you?
5 min · 12 questions · Free · Get your archetype + top 3 matching tracks
Take the quiz →