Last updated September 27, 2026.
For grading handwritten math, the strongest pick here is Grade Direct, the tool our directory still lists under its former name, Graded Pro. It reads handwritten working, including equations, diagrams and annotations, transcribes it and grades it against your rubric (how it works, checked September 27, 2026). Two things change that recommendation. If your district already pays for Gradescope, its AI grading groups a class set of identical methods faster than anything else on this list, though that feature sits behind the institutional license and not the free plan. And if the job is tutoring rather than grading, Khanmigo is free for teachers, with a caveat from new research that matters more than any feature comparison.
Six tools are set out below, each against the specific job it does: grading at scale, grading handwriting, grading on a budget, tutoring misconceptions, funded intervention, and turning a photographed equation into something you can edit.
Three things turned up in this September check that are worth naming before the list, because two of them are corrections to what this page used to say.
Gradescope's answer grouping and AI-assisted grading, the features that make it worth using for math, require the institutional license. The free Basic plan explicitly excludes AI features, collaborative grading, bubble sheets and LMS integration (pricing, checked September 27, 2026). A teacher who signs up alone gets the digitizing, not the time saving.
Two vendors have moved. Graded Pro now redirects to grade.direct, and Miss Formula has moved from imgtoformula.com to missformula.com. Both directory records still carry the old addresses. One of those records also disagrees with the vendor about third-party data sharing, which the Grade Direct section sets out.
And the most-recommended AI math tutor has now been through a two-year randomized trial. It did not find what this page previously claimed. That section is below, and it is the part worth reading if you read nothing else.
The six tools at a glance
| Tool | Best for | Free tier | Paid price (checked September 27, 2026) | Privacy notes |
|---|---|---|---|---|
| Grade Direct | Grading handwritten working | 150 credits to start, 50 a month after | $25 for 6,000 credits, no expiry | Vendor states UK hosting; our record lists third-party data sharing |
| Gradescope | Grading a class set at scale | Basic plan, no AI features | Institutional license, quote only | Our record lists every privacy field as unknown |
| GradingPal | Grading on a small budget | 50 submissions a month | $14 a month for 250, $26 for unlimited | Vendor states US hosting, no third-party sharing |
| Khanmigo | Tutoring through misconceptions | Free for teachers | $4 a month for a learner or parent account | Our record lists every privacy field as unknown |
| Skye | Funded one-to-one intervention | None | From GBP 3,500 a year per school | Vendor states regional hosting, no third-party sharing |
| Miss Formula | Photographed math into editable notation | Separate free equation editor only | $3.99 a month for formulas, $18.90 for documents | Vendor states no student data collected |
Three questions that pick your tool
Work through these in order. Most math teachers stop at the first or second.
Does your school already license one of these? Gradescope in particular is bought at department or district level, and the features worth having are the ones the license unlocks. Ask before you buy anything yourself. The answer changes the whole shortlist.
Is the work handwritten? This is the question that separates math from every other subject. A tool that grades typed prose well may not read a student's working at all. Grade Direct and GradingPal both claim to read handwriting; Miss Formula exists purely to turn it into editable notation. A general chatbot does not reliably read a page of algebra, whatever else it can do.
Are you grading, or are you teaching? Grading tools return time. Tutoring tools change what a student does when they are stuck. They solve different problems and the evidence for them is very different in quality, which the research section below sets out.
Grade Direct: grading handwritten math
Grade Direct is the best fit for a math teacher grading handwritten working, because it is built to read the working rather than only the final answer. Students photograph or scan their page, the tool transcribes the handwriting, including equations and diagrams, and grades it against the rubric you set. Our directory records a rating of 5.00 from 4 reviews, which is a strong score on a small number of reviews.
Watch out for: the vendor has moved from graded.pro to grade.direct, and our directory record still points at the old address. Our record also reads yes against sharing data with third parties, the only tool here where it does, but the vendor's own policy does not support that reading. Check it yourself before you put a class set through it.
Pricing: the free plan starts with 150 credits and adds 50 a month. A short task costs about 3 credits and a full exam paper about 30, so the monthly allowance is enough to test the tool rather than to run a department on. Teacher Pro is $25 for 6,000 credits with no expiry, which works out at roughly 200 full papers for a one-time payment. School accounts pool credits across teachers at custom pricing (pricing and credits, checked September 27, 2026).
Safety: our directory record lists UK data hosting and states yes against GDPR, COPPA and FERPA, and yes against both collecting student data and sharing it with third parties. Those entries are the vendor's own declarations to our directory, not an audit, and on the last of them our record and the vendor disagree. The privacy policy states that Grade Direct does not sell, rent or lease personal data to any third party, and shares it only with named sub-processors, OpenAI, DigitalOcean, Google, Stripe and Mailgun, to deliver the service. Sub-processors are not the same thing as third-party sharing, and a district reviewing this tool should read that list rather than our yes. A US school will still need its own data agreement.
Gradescope: grading a class set at scale
Gradescope is the best fit for grading a large set of the same assessment, because its AI grading clusters identical solutions so you grade one method and apply it to everyone who used it. With 150 scripts and three recurring misconceptions, nothing else here saves as much time.
Watch out for: the feature that does this is not in the free plan. Gradescope Basic is free after a one-term institutional trial, and it excludes AI-powered grading, AI-powered roster matching, collaborative grading, anonymous grading, programming and bubble sheet assignments, LMS integration and SSO. If you sign up as an individual teacher expecting the grouping, you will not get it.
Pricing: no figures are published. The institutional plan is quote only, and the Basic plan is free with the exclusions above (pricing, checked September 27, 2026). Gradescope's parent company is Turnitin.
Safety: our directory record lists every one of its six privacy fields as unknown, including hosting location, minimum age, and whether student data is used for training. That is a gap in our record rather than a finding against the product, and it means a district has to work from Gradescope's own privacy policy and its Turnitin contract rather than from anything we can tell you.
GradingPal: grading on a small budget
GradingPal is the best fit for a teacher paying out of pocket, because its free tier is the most generous of the grading tools here for steady weekly volume: 50 submissions a month, with rubric scoring and editable feedback rather than a crippled trial.
Watch out for: it is a general K-12 grading tool, not a math specialist. It reads scanned work and scores against a rubric, but nothing in the vendor's description claims the mathematical notation handling that Grade Direct and Miss Formula are built around. For word problems and explanations it is a good fit; for pages of algebraic working, test it before you rely on it.
Pricing: free for 50 submissions a month. Lite is $14 a month or $168 a year for 250 submissions a month, Pro is $26 a month or $312 a year for unlimited submissions, and school and district pricing is by quote (pricing, checked September 27, 2026).
Safety: our directory record lists US hosting, no minimum age, yes against FERPA and COPPA, GDPR in progress, collects student data, and does not share it with third parties. Those are the vendor's declarations. Canvas and Schoology integrations sit at school plan level, which matters if your district requires rostering through the LMS. The privacy policy is the document to take to your district.
Khanmigo: tutoring through misconceptions
Khanmigo is the best fit for giving students something to work with when they are stuck outside your lesson, because it is free for teachers and it is configured to coach rather than to answer. Our directory records a rating of 4.40 from 5 reviews.
Watch out for: engagement, not capability. The research below found that students who had it rarely used it at the moment it would have helped. Handing out access is not the intervention; building the habit of asking it is, and that is teacher work.
Pricing: free for teachers, including lesson planning, rubrics and student progress summaries. A learner or parent account is $4 a month or $44 a year. District pricing, which adds analytics, SSO and professional development, is by request (pricing, checked September 27, 2026).
Safety: as with Gradescope, our directory record lists all six privacy fields as unknown, so the Khan Academy privacy policy is the primary document. Khan Academy is a nonprofit, which shapes its incentives differently from the commercial tools here, but that is not itself a privacy guarantee and should not be treated as one.
Skye: funded one-to-one intervention
Skye, from Third Space Learning, is the best fit for a school with an intervention budget and a group of students who need spoken one-to-one math support, because it is the only tool here sold as a tutoring program rather than as software you operate.
Watch out for: the published pricing is UK pricing, in pounds, and is organized around English year groups. Third Space Learning runs a US site, but no dollar figures appear on the pricing page, so a US school has to ask. Treat the figures below as an indication of scale, not as a quote.
Pricing: fixed annual pricing by school size rather than per student. One-form entry primary is GBP 3,500 a year, two-form GBP 5,000, three-form GBP 6,000, all covering Years 3 to 6. Secondary schools are GBP 5,000 a year for Year 7 and GCSE. Sessions and student numbers are unlimited within that fee, with discounts for long bookings and multi-academy trusts (pricing, checked September 27, 2026).
Safety: our directory record gives this tool the fullest privacy entry of the six. It lists a minimum age of 8, states yes against GDPR, COPPA and FERPA, records that student data is not shared with third parties, and specifies hosting by region: UK student data in the UK or EU, US student data in the US. Regional hosting is the detail a district data officer will ask about first, and it is the reason this record is the most complete here. The privacy policy sets out the rest.
Miss Formula: photographed math into editable notation
Miss Formula is the best fit for the specific job of turning an image of an equation into something you can edit, because that is the only thing it does. Photograph a page from a textbook or your own board work and it returns editable Word equations, LaTeX, Markdown or Excel. For building worksheets and answer keys that carry real notation, it removes a job that otherwise eats an evening.
Watch out for: it has moved from imgtoformula.com to missformula.com, and our directory record still lists the old domain. Its terms set a minimum age of 18 and our record lists it against higher education and adult learners, so this is a tool for you, not for your students.
Pricing: the formula-only plan is $3.99 a month for 5,000 credits or $39.90 a year for 60,000. The full plan, which adds document and PDF conversion, is $18.90 a month for 2,000 AI credits or $100 a year for 20,000. A formula costs 1 credit and a document page costs 10. The vendor publishes a separate free LaTeX equation editor but states no free monthly allowance for the conversion service itself (pricing, checked September 27, 2026).
Safety: our record lists US hosting, states GDPR yes, and records that it collects no student data, which follows from it being a teacher tool. The privacy policy is short.
What the research actually found about AI math tutoring
This page used to say that Khan Academy's own research reported meaningfully stronger gains for students using Khanmigo, particularly those who usually get left behind. That claim carried no source and it does not survive contact with the evidence published since. It has been removed, and here is what replaced it.
In August 2026, Philip Oreopoulos and Nina Low published One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment through Brown University's Annenberg Institute. It is a two-year cluster randomized trial across 18 Tennessee middle schools, with students using Khan Academy and Khanmigo, configured to coach rather than to answer, during daily remedial math sessions.
The findings split cleanly in two.
On achievement, the effect was real and small. Math achievement rose by 1.3 national percentile ranks per term, which the authors put at roughly 0.06 to 0.08 standard deviations over a school year. For students who actively participated all year, the implied effect reached 0.14 standard deviations.
On use, the picture is the one that should change what you do. Ninety-six percent of students tried Khanmigo at least once, but the median student messaged it on only a third of the days they practiced. In sessions where a student made a mistake, the moment the tutor exists for, they engaged it in 17 percent of cases. Most messages were bare answers or clicks on a suggested prompt.
Reporting on the study, Chalkbeat quoted Oreopoulos directly: "If a keen, engaged student wants to use the technology as a good tutor, it's ready to go. The issue is that most students don't use it that way." The researchers' own conclusion was that what limited the results was engagement rather than the technology, and that human attention may be the ingredient that makes personalized learning pay (Chalkbeat, August 25, 2026).
Our view is that this is the most useful finding available to a math department this year, and that it points at a teaching decision rather than a purchasing one. An AI tutor that a student opens at the moment they are stuck is worth having. One that sits unopened in a tab is worth nothing, and no amount of licensing fixes that. If you adopt a tutor, the work is in the routine you build around it: naming the moment students should ask it, and checking that they did.
It is also a useful corrective to the wider claim that AI tutoring transforms outcomes. Two years, 18 schools, and the result is a small positive effect driven mostly by students who were going to engage anyway.
The big names, briefly
Teachers ask about these, and AI assistants recommend them, so here is where each one sits for math specifically.
ChatGPT is useful for drafting problem sets, explanations and unit plans, and unreliable as an arithmetic engine unless you have it write and run code. Check every worked solution before it reaches a student.
Gemini Notebook, the tool Google renamed from NotebookLM in July 2026, is strong at summarizing and questioning documents you give it, which makes it good for curriculum maps and exam board specifications rather than for grading or computation.
MagicSchool covers a wide spread of teacher tasks with math among them, and is a reasonable first stop if you want one account rather than six, though nothing in it is built for reading handwritten working.
Diffit turns a text or a topic into leveled materials, which serves word problems and math reading well and does nothing for notation.
Quizlet remains a study and recall tool. For math it drills vocabulary and facts, not method.
None of these five is built to read a student's algebra off a page. That is the gap the specialist tools in this article exist to fill.
A plan for one week
This is a plan to follow, not an account of a trial that has already happened. It is written for a teacher of four math classes who has never used any of these.
Before you start, ask your department head or district technology lead one question: do we already have Gradescope, or anything else on this list? If the answer is yes, stop here and use what is paid for.
Monday. Pick one class and one assessment you have already graded by hand. Scan or photograph ten scripts. Put five through Grade Direct's free credits and five through GradingPal's free tier, with the same rubric in both.
Tuesday. Compare both sets against your own grading. You are looking for one thing: did the tool read the working, or only the answer? Where a student made a method error and still reached the right answer, did it notice? That single case tells you more than any feature list.
Wednesday. Take the misconception that came up most often and ask Khanmigo to tutor a student through it. Watch what it does when the student gives a wrong answer. Decide whether you would be content for a student to meet that explanation without you there.
Thursday. If you build your own worksheets, photograph one page of notation you would otherwise retype and run it through Miss Formula. Time both routes honestly.
Friday. Add up the minutes and decide. One tool, adopted properly, beats four trialed and abandoned. If nothing saved real time, that is a legitimate result and worth writing down for the next person in your department who asks.
The choice you are making is narrower than it looks. Grading tools buy back evenings. Tutoring tools change what happens when a student is stuck, and the evidence says that depends on whether they ask. Pick for the job you actually have, check the price against the free tier before you commit, and take the privacy policy to your district rather than the marketing page.
For the wider grading field beyond math, see our guide to AI grading tools for teachers. For AI tutors chosen from the student's side rather than the teacher's, see AI tutors for students. For math support in special education, see special education math software. And once you are grading digitally, the data those tools generate is only useful if you read it, which is the subject of AI tools for analyzing student data.
