Cycling through all six. Tap any point to stop.

Measured on six things

Our French AI math assistant

Speed

Desktop Lighthouse lands at 86/100, but mobile drags at 65/100 with a 5.5s LCP and 337ms TBT. Layout is rock-solid (CLS 0.002) — it's the first paint that's slow on phones.

Screenshots

1 / 1

MathGPT

About this app

MathGPT is our French-language AI math assistant. Submit a question as text or a photo and read step-by-step explanations. The workspace includes PDF support, quizzes, calculators, and a math keyboard. It is designed to help students understand their work and continue with follow-up questions.

Key features

1Photo, text and PDF input for math problems, with step-by-step explanations written in French.
2Four subject modes — Mathématiques, Physique, Chimie, Biologie — sharing the same photo/PDF workspace.
3Revision toolkit: quizzes generated from chapters or documents, calculators, function plotter and solution history.

Target customer profile

Drawn from the product itself, not from a survey.

Scout Review

Written by AI from measured evidence, scored out of 100.

MathGPT is a French-language AI math assistant that takes a question as text, photo or PDF and returns step-by-step explanations, wrapped in a workspace with quizzes, calculators and a function plotter. The execution is competent across the board: a clean, well-organized interface, a 97/100 accessibility audit, a perfect 12/12 security header checklist, and a desktop performance score of 86. The weak points are equally clear. Mobile performance sits at 65/100 with a 5.5-second LCP, which matters for a product students will use on phones. First-load clutter — a cookie modal over the upload zone and a promo card over the sidebar — undercuts an otherwise tidy design. And there's no demo, no trial, and no published compliance or accessibility documentation, so a buyer has to take the pitch on faith. The product's real bet is the French-language, step-by-step pedagogy plus the PDF and revision workflow. That's a legitimate niche, but it's competing against free general-purpose chatbots that also answer in French, so the differentiation has to be demonstrably better, not just differently packaged.

Measure by measure

Usability

One question box, one upload zone, four subject tabs — the core flow is obvious in seconds. Onboarding is thin though: no demo, no trial, and 'Mode avancé' hides in low contrast.

Design

Clean light UI with a coherent infinity logo and blue accent system. Sidebar and subject tabs read clearly, but the cookie modal and floating promo card crowd the first-load view.

Speed

Desktop Lighthouse lands at 86/100, but mobile drags at 65/100 with a 5.5s LCP and 337ms TBT. Layout is rock-solid (CLS 0.002) — it's the first paint that's slow on phones.

Security

Full 12/12 header checklist: HTTPS enforced, HSTS, CSP, frame protection, Referrer-Policy, Permissions-Policy, plus SPF and DMARC. No published audit or bug bounty found.

Accessibility

Lighthouse accessibility 97/100 with a single contrast failure. French-only, no a11y statement, but the automated baseline is genuinely strong.

Growth

A real, specific wedge: French-language step-by-step math help with photo and PDF input for collège-to-university students. Differentiated on language and workflow, not on the underlying model.

Improvement areas

Mobile performance: 65/100 Lighthouse with a 5.5s LCP is slow for a phone-first student audience.
First-load clutter: cookie modal covers the upload zone and the floating promo card overlaps sidebar navigation.
No demo, trial or published data-handling detail to help a visitor decide before paying.

Design & Usability

MathGPT's interface is genuinely well-organized for its job. The homepage opens on a single bordered question box with a clear 'Pose ta question' placeholder, a dashed photo/PDF upload zone beneath it, and four subject tabs (Mathématiques, Physique, Chimie, Biologie) across the top. The sidebar gives quick access to Nouvelle question, PDF, Quiz, Calculatrices and Traceur de fonctions, all with consistent blue icons and French labels. The infinity logo and blue accent color carry through cleanly. The problems are all first-load clutter: a cookie consent modal sits directly over the upload area, and a floating yellow 'Découvrir MathGPT Pro' card overlaps the sidebar's bottom edge. Subject tabs also use a subtle gray that blends into the page compared to the strong blue CTAs. Usability-wise, the core flow is obvious within seconds — type, upload, or paste, then send. But onboarding is thin: there's no live demo, no free trial described, and the 'Mode avancé' toggle sits in low contrast inside the input box. The upload constraints (8 Mo photos, 20 Mo PDFs) are stated plainly, which is a nice touch. For a student landing cold, the path to value is short; for a student deciding whether to pay, the page gives them little to go on.

Speed & Security

Speed is a split verdict. Desktop Lighthouse performance lands at 86/100, which is solid, and the layout is exceptionally stable at CLS 0.002 — nothing jumps around while you read. Mobile is the weak spot: 65/100 with a Largest Contentful Paint of 5.5 seconds, First Contentful Paint at 3.3s and Total Blocking Time of 337ms. On a phone, that's a noticeably slow first impression for a tool students will likely reach for on exactly that device. Security is the strongest dimension here. The HTTP probe returned a perfect 12/12: HTTPS enforced, HSTS, Content-Security-Policy, frame protection, X-Content-Type-Options, Referrer-Policy and Permissions-Policy all present, plus SPF and DMARC records on the domain. That's a more complete header posture than many early-stage SaaS products bother with. What's missing is the paperwork layer — no SOC 2, ISO 27001, penetration test summary or bug bounty program appears anywhere in the collected content. For a consumer education tool handling student homework and uploaded PDFs, that's not unusual at this stage, but it does mean privacy-conscious parents or schools have nothing formal to point at.

Accessibility & Growth

Accessibility is a genuine strength. Lighthouse's axe-core audit scored 97/100, with exactly one failing check: insufficient background/foreground color contrast. That's a fixable, narrow issue rather than a systemic one, and it's the kind of thing a single CSS pass resolves. Beyond the automated score, the site is French-only with no language switcher, and no accessibility statement or WCAG conformance claim appears in the collected content — normal for a product this young, but worth noting for institutional buyers. Growth is where the interesting question sits. The idea is real and specific: French-language AI math assistance with step-by-step reasoning, photo input, PDF workspace, quizzes, calculators and a function plotter, aimed at students from collège through the first years of higher education. That's a coherent wedge. The French-language angle is genuinely underserved — most AI math tools default to English and translate poorly on mathematical vocabulary. The workflow bundling (photo → explanation → follow-up → quiz → history) is a real product thesis, not a feature list. The honest limit: the underlying capability is a commodity. ChatGPT, Gemini and Claude all solve math in French for free, and the site's own pitch — 'moins qu'une heure de cours particulier' — competes against a free alternative, not just against tutors. The moat has to be the French pedagogy, the PDF workspace and the revision loop, and those need to be visibly better than a chatbot prompt to justify a subscription.

Conclusion

If you're a French-speaking student who wants explanations in your own language with a revision loop attached, MathGPT is a reasonable place to start — the core flow is fast to learn and the accessibility and security fundamentals are solidly handled. If you're evaluating it as a paid tool, the case is thinner: the site doesn't show you a demo, doesn't offer a trial, and doesn't publish anything about data handling beyond a privacy policy. The single most valuable thing the team could do next is make the difference from a general chatbot visible on the homepage — show a real French exercise, show the step-by-step reasoning, show the quiz generated from the same PDF. Right now the pitch is clear but the proof is missing. Fix the mobile load time, get the cookie banner out of the upload zone, and give visitors one honest before-and-after, and this becomes a much easier recommendation.

Suggestions for improvement

Cut mobile LCP from 5.5s toward 2.5s — defer non-critical JS and optimize the hero render path.
Move the cookie consent modal off the upload zone and reposition the floating Pro promo card so it stops overlapping the sidebar.
Add a live demo or a short free trial so visitors can see a real step-by-step solution before hitting the paywall.
Publish an accessibility statement and a short data-handling page covering how uploaded photos and PDFs are stored and deleted.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	MathGPT
Language of explanations	French-first, step-by-step reasoning written for French curricula	Not specified in provided data
Input methods	Text, photo (up to 8 Mo), PDF (up to 20 Mo), math keyboard	Not specified in provided data
Subject coverage	Mathématiques, Physique, Chimie, Biologie	Not specified in provided data
Revision and practice tools	Quiz generation, calculators, function plotter, solution history	Not specified in provided data
Free trial or demo	No demo or trial described on the site; Pro upsell present	Not specified in provided data
Published security and compliance	Full HTTP security headers (12/12); no SOC 2, ISO or bug bounty published	Not specified in provided data