Marksy vs MyRevisionAgent
Both say about 92% of IAs land within one grade of the IB. Here is what that number hides, and what happened when we sent each one the same essay.
Based on what each tool says publicly, checked on (MyRevisionAgent methodology page, MyRevisionAgent IA grader page, MyRevisionAgent EE grader page). Prices and features change, so check before you pay.
Same headline, different evidence
MyRevisionAgent's figures are what it publishes on its own site. We can't check them, because it doesn't share its test essays.
| What | Marksy | MyRevisionAgent |
|---|---|---|
| IAs within one grade | 92% on 1,317 IAs and TOK pieces | 92% on 1,678 IAs |
| Exact grade | 42% | 43% |
| Average marks off | 2.0 | About 2.3 |
| Compared with guessing the average | Yes: guessing gets 79% when every grade counts, Marksy 88% | Not published |
| Results for weak and strong essays | Shown for every grade from 3 to 7 | Not published |
| Compared with chatbots | ChatGPT, Claude and Gemini on the same 226 essays, prompt described | GPT-5.5 and Claude Opus 4.8, prompts not published |
| Where the answer marks come from | Essays students shared with their IB moderated mark | Described as examiner-marked, source not stated |
| Extended Essay | Not tested yet | 96% within one band on the criteria used until November 2026. Nothing published for the May 2027 criteria |
| TOK | Tested: TOK essay and exhibition | No accuracy figure published |
| Price | 5 free gradings a month, then monthly plans | £9.99 per grade after 2 free |
Why 92% on its own proves little
Most IAs score a 5 or a 6. A grader that never reads the essay and gives everyone their subject's average mark still lands within one grade about 89% of the time. So a 92% headline can't tell you whether a grader actually reads your work.
The fix is to count every grade equally, so the 3s and 7s matter as much as the 5s. That guess then drops to 79%. Marksy holds 88%. MyRevisionAgent doesn't publish either number.
We sent both the same essay with nothing right in it
A 3,200-word Business Management Extended Essay that looks the part: contents page, long method section, six tables, citations everywhere, and SWOT, Porter, Ansoff and BCG all named. Underneath, every theory is defined wrongly, market share is measured by counting car-park spaces, and the headline result is an average of unrelated numbers. A fair mark is about 2 to 5 out of 34.
Its own summary says the essay fails
Your tools are mis-defined and your methods measure the wrong things, so every conclusion, including 73.4%, rests on invalid evidence.
Then it gives the essay a B. Its note at the bottom explains why:
Cross-checked: two independent assessments of this essay differed by 16.5 points. Your grade comes from whichever of the two our evaluation shows is more accurate, not an average of them.
What this suggests
MyRevisionAgent says an AI reads the essay and scores features of it, and a trained model turns those scores into the mark. Here one assessment saw the problems and the other gave 16.5 more marks, and the higher one won. A formula that rewards structure and citations can be fooled by a tidy draft that is wrong underneath.
One essay doesn't prove a rule, and our 7/30 is a little generous too. That's why we also test on 226 real essays with the same number of weak, middle and strong ones. On that set Marksy scores 74/100 on our marking score. ChatGPT scores 65 and Claude 61.
See how we test accuracyTested on 2 November 2026 with the same PDF and research question. MyRevisionAgent was set to the November 2026 session (marked out of 34). Marksy uses the May 2027 criteria (marked out of 30).
Questions people ask
Check your draft before you submit it
5 free gradings a month, with a mark for every criterion and comments on your own sentences.