Education Custom GPTs
Last reviewed
May 13, 2026
Sources
40 citations
Review status
Source-backed
Revision
v2 ยท 3,996 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
May 13, 2026
Sources
40 citations
Review status
Source-backed
Revision
v2 ยท 3,996 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Custom GPTs, GPT Store, ChatGPT, AI in Education, and Education
Education Custom GPTs are the Custom GPTs listed in the Education section of OpenAI's GPT Store. Education was one of the eight original sections that OpenAI used to organize the store when it launched on January 10, 2024, alongside Top Picks, DALL-E, Writing, Productivity, Research and Analysis, Programming, and Lifestyle. The official description that appears on the section header reads "Explore new ideas, revisit existing skills," and the category collects GPTs that help users learn a subject, work through homework, summarize lectures or papers, draft and revise essays, or prepare for exams.
Many GPTs in this section come from established education organizations. Khan Academy ships Code Tutor, a coding spinoff of its Khanmigo tutor, and CK-12 publishes a version of its Flexi tutor for middle and high school math and science. Independent builders dominate the rest of the rankings: Universal Primer by investor Siqi Chen, AlphaNotes for YouTube and article summaries, Math Solver, Scholar GPT, Consensus, and the Wolfram GPT. Beyond the store, OpenAI built two larger education programs around the same product: ChatGPT Edu, a per-seat plan for universities announced May 30, 2024, and the OpenAI Academy, a free AI literacy hub launched in September 2024.
The Education section of the GPT Store groups GPTs whose primary use case is teaching, tutoring, study support, or academic research. Builders self-select a category when they publish a GPT, and OpenAI's editorial team can promote individual GPTs to the front of the section. The ranking is largely driven by total chats, so consumer-facing study aids tend to outrank niche research GPTs.
Education GPTs typically combine three building blocks: a long system prompt that fixes the tutor's behavior (often a Socratic style that asks rather than answers), uploaded knowledge files (textbook chapters, problem sets, course notes), and Actions that call external APIs (Wolfram for symbolic math, Consensus or ScholarAI for paper search, YouTube transcript fetchers for video summaries). The same underlying model powers all of them; the personality, retrieval, and tool calls are what differentiate Code Tutor from a generic chat with GPT-4o.
This article covers the GPT Store Education category and its most popular members, the broader OpenAI education programs (ChatGPT Edu and OpenAI Academy), Khan Academy's separate Khanmigo product, Anthropic's competing Claude for Education offering, the pedagogical research vendors cite, common use cases, and recurring concerns around cheating, hallucination, and biased AI-text detection.
The GPT Store opened on January 10, 2024, two months after Sam Altman demonstrated Custom GPTs at OpenAI's first DevDay on November 6, 2023. At launch, OpenAI said users had already built more than 3 million custom GPTs and that the store contained roughly 159,000 published ones. The store was initially limited to paying subscribers (Plus at $20 per month, Team at $25 per user per month, and Enterprise); free users gained access in May 2024 alongside the GPT-4o release.
Education GPTs were a launch marketing focus. The blog post highlighted Khan Academy's Code Tutor and the CK-12 Flexi tutor by name as examples of high-quality GPTs the store was meant to showcase, alongside the Top Picks, DALL-E, Writing, Productivity, Research and Analysis, Programming, and Lifestyle rows.
Each section on the store lists GPTs in three ways:
| View | What it shows | How it is ordered |
|---|---|---|
| Featured | OpenAI editorial picks | Curated by OpenAI staff |
| Trending | GPTs gaining traction in the section | Recent growth in chats |
| Top by category | Most-used GPTs in the category | Total chats since publication |
In the Education section, the Featured row at launch included Code Tutor by Khan Academy and CK-12 Flexi, and the trending and top-by-category rows have been dominated since 2024 by a mix of homework helpers (Math Solver, Universal Primer), summarizers (AlphaNotes), and research tools (Consensus, Scholar GPT, ScholarAI, Wolfram).
The table below lists Education GPTs that have been independently documented as appearing in the GPT Store's Education category and that show up on third-party leaderboards such as gptstore.ai and featuredgpts.com. Total-chat counts are not published by OpenAI; the entries below are ordered roughly by sustained visibility on the Education section's Top and Trending rows since January 2024.
| GPT | Builder | What it does |
|---|---|---|
| Universal Primer | Siqi Chen | Recursively teaches any topic by first checking prerequisite concepts and using analogies, written by a16z scout and Runway investor Siqi Chen. Promoted by Chen on X and LinkedIn as the "#1 education GPT" in the store in early 2024. |
| Code Tutor | Khan Academy | A Khanmigo-style Socratic coding tutor that refuses to write code outright and instead asks guiding questions. Featured on the GPT Store launch home page on January 10, 2024. |
| CK-12 Flexi | CK-12 Foundation | Middle and high school math and science tutor based on the CK-12 OER textbook library. Also featured at GPT Store launch. |
| AlphaNotes | AlphaNotes (alphanotes.one) | Summarizes YouTube videos and long articles into structured study notes and quizzes, callable in any chat with @alphanotes. |
| Math Solver | Independent | Solves and explains math problems step by step, from arithmetic up through calculus and linear algebra. |
| Wolfram GPT | Wolfram Research | Connects ChatGPT to Wolfram|Alpha and the Wolfram Language for symbolic math, curated data, and visualisation; one of the first plugins announced for ChatGPT in March 2023 and ported to a Custom GPT after the store launched. |
| Consensus | Consensus (consensus.app) | Searches and synthesizes findings from a 200M-paper academic index, returning answers with linked citations. Founded by Eric Olson and Christian Salem. |
| Scholar GPT | awesomegpts.ai | Connects to Google Scholar, PubMed, bioRxiv, and arXiv for literature search and citation generation. |
| ScholarAI | ScholarAI (scholarai.io) | Searches and summarizes peer-reviewed papers, patents, and books across PubMed, arXiv, and bioRxiv. |
| Khanmigo Lite | Khan Academy | A trimmed-down public version of Khan Academy's full Khanmigo tutor that ships in the store. |
| Mr. Ranedeer AI Tutor | Independent (Jushua) | Adaptive personalized lesson planner that asks for the learner's level and learning style before each lesson. |
A few caveats. OpenAI does not publish authoritative usage figures, so third-party rankings are best-effort scrapes. Names like "Math Solver" and "Scholar GPT" are popular enough that several unrelated GPTs use the same display name; only the GPT URLs identify a specific publisher. OpenAI also removes GPTs that violate trademark or content policy, so individual entries can disappear without notice.
OpenAI announced ChatGPT Edu on May 30, 2024, as a dedicated ChatGPT plan for colleges and universities. The Edu plan packages GPT-4o access with the same controls that ship in ChatGPT Enterprise (single sign-on, group permissions, no training on customer data, SOC 2 compliance, data retention controls) at a price negotiated per institution. OpenAI markets Edu as a way for a school to give every student, faculty member, and staff member an account without the per-user friction of standard Plus subscriptions.
The announcement was framed around five early adopters whose deployments had been running since 2023 and 2024:
Later customers include the California State University system (signed February 2025 for around 460,000 students and 63,000 faculty across 23 campuses), Texas A&M, and Miami Dade College.
OpenAI does not publish a public per-seat price for the Edu plan; institutions are quoted a contracted rate after speaking to sales. For comparison, the broader ChatGPT pricing tiers that students and educators most often use directly are:
| Plan | Price | Audience |
|---|---|---|
| Free | $0 | Anyone, with rate-limited access to GPT-4o, GPT-4o mini, and the GPT Store |
| Plus | $20/month | Individual users who want higher rate limits, voice, advanced data analysis, and image generation |
| Pro | $200/month | Power users who want unlimited use of OpenAI's most capable models |
| Team | $25/user/month | Small organizations |
| Edu | Per-seat, negotiated | Accredited colleges and universities |
| Enterprise | Per-seat, negotiated | Companies and large institutions |
OpenAI also ran a temporary promotion that gave US and Canadian college students free ChatGPT Plus access from March 31 through May 31, 2025; the offer was not renewed.
The OpenAI Academy launched in September 2024 as a free hub at academy.openai.com for AI literacy content. It hosts on-demand lessons aimed at educators, students, jobseekers, small business owners, and non-profit leaders, with materials from partner organisations such as Common Sense Media and Code.org. The Academy is not part of the GPT Store; it is a separate education site run by OpenAI's global affairs team.
The most concrete output from the Common Sense partnership is ChatGPT Foundations for K-12 Educators, a free one-hour, eight-lesson course launched November 20, 2024, covering AI fundamentals, data privacy, ethical use, student safety, and example classroom workflows. Common Sense and OpenAI first announced the broader collaboration in January 2024.
Khan Academy's Khanmigo is the most visible AI tutor outside of OpenAI's first-party ecosystem, and it is also the lineage that produced Code Tutor in the GPT Store. Khan Academy released Khanmigo in a limited pilot on March 14, 2023, the same day OpenAI announced GPT-4. The original pilot was paid (a $20 per month donation to Khan Academy) and limited to a small set of US districts and donor families.
Khanmigo's defining design choice is its Socratic prompt: instead of solving a problem when a student asks, it asks the student what they think the first step should be, offers hints, then walks back through wrong turns when the student gets stuck. Khan Academy reused the same prompt template when it shipped Khanmigo Lite (also called Code Tutor in the GPT Store).
On May 21, 2024, Microsoft and Khan Academy announced a partnership that made Khanmigo for Teachers free for all US K-12 teachers, with Azure providing the infrastructure. The deal was extended to 44 additional countries in August 2024 and translated into 34 additional languages by December 2024. Microsoft and Khan Academy also said they would collaborate on an open-source math model based on the Phi-3 family of small language models. The teacher product previously cost $4 per month; the Microsoft donation eliminated the fee.
Khanmigo has also been adopted at the state level. New Hampshire became the first state to sign a statewide contract in June 2024, making Khanmigo available at no cost to students and teachers across the state. Arizona funded a $1.5 million deployment that has grown to more than 170,000 students; Arizona has the highest Khanmigo usage per capita of any US state. Reported student numbers across all Khanmigo channels rose from roughly 68,000 in the 2023-24 school year to more than 700,000 in 2024-25.
Anthropic announced Claude for Education on April 2, 2025, as the closest competitor to ChatGPT Edu. The plan gives institutions campus-wide access to Claude with a new "Learning" mode that biases the model toward Socratic dialogue rather than direct answers, an explicit nod to the pedagogical pattern that Khanmigo, Code Tutor, and Universal Primer all share.
The initial launch partners were:
| Institution | Notes |
|---|---|
| Northeastern University | Campus-wide rollout covering 50,000 students, faculty, and staff across 13 global campuses |
| London School of Economics and Political Science | Campus-wide deployment focused on equity, ethics, and skills development |
| Champlain College | Campus-wide rollout aimed at career-focused workforce skills |
Anthropic also opened Claude Pro accounts to anyone with a .edu email at the same time. The launch happened roughly eleven months after ChatGPT Edu and was widely read as Anthropic trying to peel off universities uneasy with a single-vendor commitment to OpenAI.
The sales pitch for every Education GPT and every campus AI deployment leans on a small body of educational psychology research. The two papers that tutoring vendors cite most often are Benjamin Bloom's 1984 "2 Sigma" paper and Kurt VanLehn's 2011 review.
In 1984, educational psychologist Benjamin Bloom published "The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring" in Educational Researcher. Bloom summarized two University of Chicago dissertation studies that found students tutored one-on-one with mastery learning outperformed students in regular classrooms by about two standard deviations, meaning the average tutored student scored above 98 percent of the classroom students. Bloom called one-to-one tutoring "too costly for most societies to bear on a large scale" and posed the "problem" as whether group instruction could be made nearly as effective.
The 2 Sigma figure is widely cited but worth qualifying. Bloom's famous illustration was a hand-drawn diagram, not a fit to data, and later analyses suggest the real effect of one-on-one tutoring is closer to 0.4 to 0.6 standard deviations once you control for the higher mastery thresholds Bloom's tutored groups had to meet.
In 2011, computer scientist Kurt VanLehn of Arizona State published "The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems" in Educational Psychologist. VanLehn compared dozens of studies and concluded that step-based and substep-based intelligent tutoring systems produced an average effect size of d = 0.76 relative to no tutoring, very close to the d = 0.79 he calculated for human tutoring. The result undercut the conventional wisdom (rooted in Bloom) that human tutors were dramatically better than software tutors, and it remains the most-cited piece of evidence in favor of computer tutoring.
In June 2023, Wharton professor Ethan Mollick and his co-author Lilach Mollick published "Assigning AI: Seven Approaches for Students, with Prompts" on SSRN and arXiv. The paper offers a pedagogical scaffold for instructors who want to assign work that uses AI rather than ban it, organised around seven roles the model can play in a class:
| Role | What the student asks the model to do |
|---|---|
| AI-tutor | Explain a concept and check understanding |
| AI-coach | Give feedback on the student's work |
| AI-mentor | Discuss long-running goals and reflection |
| AI-teammate | Collaborate with the student on a group task |
| AI-tool | Carry out a specific narrow task (drafting, formatting) |
| AI-simulator | Role-play scenarios such as a historical figure or a customer |
| AI-student | Be taught by the student, so the student has to explain clearly |
The framework is widely referenced in faculty teaching guides at US universities and was one of the explicit inspirations cited in OpenAI's ChatGPT Edu announcement. The Mollicks publish much of their work via the One Useful Thing newsletter.
Education GPTs and AI tutors tend to cluster around a small set of recurring use cases.
Khanmigo, CK-12 Flexi, and Code Tutor target the K-12 market directly. Their distinguishing feature is the refusal to give answers; instead the model asks the student to articulate the next step. Khanmigo has been formally adopted in New Hampshire and Arizona, and CK-12 Flexi maps to the same Common Core, AP, and California state standards that CK-12's free textbooks already cover.
For higher education, Universal Primer, Math Solver, and Wolfram GPT cover the homework helper niche. Math-heavy courses are still hard for plain ChatGPT to get right without a calculator, so the Wolfram integration is what makes the Wolfram GPT a perennial fixture in the Education section. Universal Primer's recursive-prerequisite design fits well for a college student who has the prereqs on paper but has forgotten the actual content.
A second tier of GPTs focuses on standardized test prep (SAT, ACT, GRE, MCAT, LSAT, AP exams). These usually combine a tight system prompt fixed to a specific test format, an uploaded question bank, and explicit scoring rubrics. They sit alongside products like Mr. Ranedeer's adaptive lesson planner.
The Education category overlaps with language learning. ChatGPT's Advanced Voice Mode (launched late 2024) and its ability to switch language mid-conversation make it useful as a conversation partner without paying a human tutor. German, Mandarin, and Spanish tutor GPTs appear in the Education section, and outside the GPT Store, Duolingo's Max tier and Babbel's Speak feature integrate GPT-4.
Math, physics, chemistry, and computer-science problem solving are a heavy use case. The pattern is usually "upload a problem, get a step-by-step solution." Wolfram, Math Solver, and various physics and chemistry tutors cover this. Essay-coach GPTs give structural and grammar feedback and rubric-aligned scoring, popular with students who previously paid Princeton Review or Kaplan for a first pass.
For researchers and graduate students, Consensus, Scholar GPT, and ScholarAI search peer-reviewed papers, summarize them, and produce draft citations. Consensus indexes over 200 million papers and was used by more than 8 million people by late 2025.
AI tutoring at scale has provoked persistent worries. The most-discussed are cheating, hallucinated facts, and the bias built into AI-text detection.
Universities have wrestled with whether AI use in coursework is cheating since ChatGPT launched in November 2022. Surveys give a consistent picture: in Wiley's 2024 instructor survey, 96 percent of instructors said they believed at least some of their students had cheated in the past year, up from 72 percent in the 2021 survey, and 47 percent of students said it was easier to cheat than the year before because of generative AI. Reported student-discipline rates for AI-related plagiarism rose from 48 percent of cases in 2022-23 to 64 percent in 2023-24. Inside Higher Ed and EdSurge have both run regular coverage of campus policy responses, which range from outright bans to mandatory "AI used" disclosure clauses in syllabi.
Most guidance for instructors now leans toward redesigning assessments (in-class essays, oral examinations, project work with a paper trail) rather than relying on detection. The Mollicks' "Assigning AI" framework is one of the most-cited examples of this approach.
Every education GPT inherits the underlying model's hallucination rate. The most studied failure modes are:
Reputable Education GPTs typically chain the model to an external retrieval source (Khanmigo to Khan Academy's curriculum, CK-12 Flexi to the CK-12 textbook library, ScholarAI to PubMed and arXiv) to suppress hallucinations on the highest-risk question types.
A widely cited Stanford study by Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou, published in Patterns in July 2023, found that commercial GPT detectors misclassify more than half of TOEFL essays written by non-native English speakers as AI-generated, while correctly identifying eighth-grade US essays. The researchers attributed the bias to lower perplexity (more predictable word choice) in non-native English writing, and they showed that simple prompt rewriting could let actual AI-written text slip past the same detectors. The paper has been used by international-student advocacy groups to push universities not to penalise students based on detector output alone. The Markup also documented specific cases of international students who were falsely accused of cheating because of detector flags.
OpenAI itself shipped a public AI-text classifier in January 2023 and withdrew it in July 2023, citing low accuracy. As of 2026, no major lab ships a detector it claims is reliable for academic disciplinary use.