Built by students,
for students.
We sat the PAT to get into New College, Oxford. We know how it feels. So we built the tool we wish we had.
Our story
We faced the challenge ourselves.
As students who studied for the PAT recently, we know how stressful exam-time can be. The biggest constraint we faced was a lack of past-paper material that was actually relevant.
Papers from pre-2016 are outdated. The syllabus has changed significantly since then and the old material reflects it.
Most PAT/MAT questions are long-form. Helpful for interviews, but not the punchy MCQ format of the modern ESAT or TMUA.
MAT questions only cover maths, and sometimes assume calculus knowledge that is not relevant for the ESAT.
Very little chemistry and biology material exists outside of NSAA past papers. That is a huge gap for ESAT candidates.
What you really need is a large, unending corpus of questions generated in the exact style and subject matter of your target exam (TMUA's pure maths and logical reasoning, or ESAT's maths, physics, chemistry and biology), drafted against every real past paper and each exam's official syllabus as a hard guideline, then checked and modified by Oxford students before it reaches you, with full worked solutions, variable difficulty, and an intelligent tutor to guide you through each one. That is what we built.
The question bank
2,000+ questions and growing.
The question bank is a live, growing ecosystem of TMUA-style and ESAT-style questions, each drafted to match the precise tone, difficulty curve, and subject scope of its target exam, then checked and, where needed, modified by Oxford students before it goes live. Every question and worked solution gets stored for instant retrieval, so anyone who encounters the same problem benefits from work already done. Gaps are detected automatically: when the system identifies under-covered topics, it drafts new questions to fill them, which are then reviewed and added to the bank, keeping it balanced across each exam's syllabus.
Nothing is locked in. If a worked solution needs correcting, users can flag it and suggest an amendment, and an Oxford student reviews the flag and either modifies the solution directly or has it regenerated. The bank improves continuously from real usage.
Trained on olympiad material
To push difficulty to the top end, the model was also trained on IMO, BMO and Physics Olympiad problems. Higher-difficulty questions draw on this to produce genuinely hard, multi-step problems, which Oxford students then check for correctness and sharpen where needed.
Answer verification
Every generated question goes through an automated post-verification step that independently solves the problem and checks the result matches the provided answer key, and a failed check is escalated to an Oxford student to fix or discard rather than being silently regenerated.
Worked solution review
An automated pass checks that the worked solution is logically consistent with both the question stem and the correct answer, catching cases where the solution explains a different route to the right answer. Oxford students spot-check and, where the explanation is muddled, rewrite it in plain language.
Distractor explanations
After answering, hover any wrong option and click "Why?" to see the specific misconception or reasoning slip that leads someone to choose it, drafted by AI and checked by Oxford students. Generated once, stored permanently, available instantly to every student who encounters that question.
Generation takes around 25 seconds per question because of these verification steps. The numbers below reflect the live state of the bank right now.
Exam landscape
What exams will I need to sit?
In 2024 the system changed significantly. Previously, STEM entrance exams ranged across the PAT, MAT, NSAA, TMUA, ENGAA and STEP. Now, all science-based ones have been consolidated into just two exams: the ESAT and the TMUA.
Discontinued as of 2024
The MAT (Mathematics Admissions Test), PAT (Physics Aptitude Test), and BMAT (BioMedical Admissions Test) have all been discontinued. Do not rely on them as your primary prep material.
Engineering & Science Admissions Test. Covers maths, physics, chemistry, and biology. Required for most STEM courses at Oxford, Cambridge, and Imperial.
We can help ✓Test of Mathematics for University Admission. Required for maths, computer science, and economics courses at Oxford, Cambridge, and Imperial.
We can help ✓Check the course page for your chosen university to confirm which test you need. A full breakdown is below.
By university
Course → exam lookup
Select your university to see which test each course requires.
| Course | Admissions Test |
|---|---|
| Mathematics | TMUA |
| Mathematics & Statistics | TMUA |
| Mathematics & Computer Science | TMUA |
| Computer Science | TMUA |
| Computer Science & Philosophy | TMUA |
| Physics | ESAT |
| Physics & Philosophy | ESAT |
| Engineering Science | ESAT |
| Chemistry | ESAT |
| Biochemistry (Molecular & Cellular) | ESAT |
| Biomedical Sciences | ESAT |
| Medicine | UCAT |
| Most non-STEM courses (previously TSA) | TARA |
| Materials Science | — |
Common question
Why not just use ChatGPT or Claude?
You can, and general models are genuinely good at maths and science. But asking one to "give me an ESAT-style question" leaves too much to chance: it has no fixed sense of the real exam's syllabus, difficulty curve, or question format, and nobody who has actually sat the exam checks what comes out before you see it. A subscription to a frontier model also still costs more than we do. We built Oxcel to remove exactly that guesswork, pairing AI drafting with Oxford students who check and, where needed, modify every question.
No wait time on 1,500+ questions
A chat conversation disappears the moment you close the tab, and a fresh question takes about 25 seconds to draft and check. Neither is a problem here: over 1,500 questions, drafted by AI and checked by Oxford students, plus every historical past paper, already sit in the bank ready to answer instantly.
Calibrated to the real exam
A general chatbot has no ground truth for what a difficulty-4 ESAT question actually looks like. Ours is trained on every real past paper for each exam and then checked by Oxford students who sat these exams, so the format, phrasing, and difficulty curve match what you will see on test day, not a generic approximation of it.
Answers you can trust
Ask ChatGPT for ten questions and you have to manually check each answer yourself, since models occasionally make arithmetic or logical slips with no way to catch them. Every question here is independently re-solved by an automated check and then reviewed, and if needed modified, by an Oxford student before it is ever shown to a student.
No degrading chats to restart
Push a frontier model through a long conversation and question quality starts to drift, with a noticeable tendency to repeat the same phrasing, numbers, or question setups the longer the chat runs, so you end up starting a fresh chat every so often just to reset it. Every question here is generated as its own independent, verified unit, so quality and variety never depend on how long your session has been running.
Built for exam prep, not chat
Timed past-paper mode, quick-fire weakness targeting, progress tracking, distractor explanations, and a whiteboard-aware tutor are all purpose-built for revision. That is a lot of exam-specific tooling to recreate by hand in a chat window every time you sit down to study.
Cheaper than any frontier model
Basic starts at £20/month, undercutting a standalone subscription to ChatGPT, Claude, or Gemini, and you get exam-specific question generation, past papers, and a tutor bundled in rather than a general-purpose chat window.
General AI is a great study companion. We just think the questions you practise on should be verified, exam-accurate, and free the moment someone else has already asked for them.
Ready to start?
2,000+ questions, every past paper, full worked solutions. Basic from £20/month · Elite from £60/month.
Get started →