What does “within one grade” mean?

The IB turns your IA mark into a grade from 1 to 7 (A to E for TOK). If the IB gave your IA a 5, Marksy is within one grade when it says 4, 5 or 6. Marksy did this for 92% of the 1,317 essays we tested.

Is Marksy more accurate than ChatGPT?

In our test, yes. We pasted the official criteria and the same 226 essays into ChatGPT, Claude and Gemini. Their marks were 18% to 33% further from the IB's than Marksy's. ChatGPT and Claude also got far more grade 7 essays wrong.

Is Marksy as accurate as my teacher?

Not yet, and it isn't a fair race. For most IAs, the IB's final mark is your teacher's mark, moved up or down for the whole class after the IB re-marks a sample. So a teacher is checked against an answer that starts from their own mark. On our test, a teacher's mark matched the IB's exact grade 68% of the time, against 42% for Marksy. Marksy is useful because you can check a draft whenever you want, not because it replaces your teacher.

Is Marksy just ChatGPT with a prompt?

No. Your draft goes through several steps, and each one has been tuned and tested against essays with real IB marks. When we pasted the official criteria and the same essays into ChatGPT, its marks were 18% further from the IB's than Marksy's.

Does this cover the Extended Essay?

Not yet. The Extended Essays we hold are almost all high-scoring, so they can't show whether a grader tells weak and strong essays apart. We will add EE results when we have a fair set.

Where do the marks you tested against come from?

Students shared their IAs and TOK pieces publicly together with the mark the IB gave them after moderation. We use those marks as the answer key. Marksy never saw the answer for an essay it was scored on.

Why is the marking score out of 100?

It puts every grader on one scale that's easy to read. 0 is what you'd get from random marks and 100 is as close as a student's own teacher. Under the hood it's the average distance from the IB's mark, with weak, middle and strong essays counted equally.

How accurate is Marksy?

We graded 1,317 real IAs and TOK essays and compared Marksy's mark with the one the IB gave. Here is how close it got, how it compares with pasting your IA into ChatGPT, Claude or Gemini, and where it still falls short.

Grade my draft How we tested

Within one grade of the IB: 92%; of essays marked within one grade of the IB's final mark
Marking score: 74/100; ChatGPT 65, Claude 61, Gemini 57
Big misses: 17 in 100; Essays 2+ grades off. Claude: 26

Marksy vs ChatGPT, Claude and Gemini

We pasted the official criteria and the same 226 essays into each one, the way you would. These 226 have the same number of weak, middle and strong essays, so no grader can score well by giving everyone a 5.

ChatGPT's marks were 18% further off than Marksy's, Claude's 24% and Gemini's 33%.
Marksy made the fewest big misses: 17 essays in 100 were two or more grades off, against 22 to 26 for the chatbots.
ChatGPT and Claude mark strong work far too low. Claude put 71% of grade 7 essays two or more grades too low.

A teacher
The student's own teacher's mark
100
Marksy Full review
Our detailed review
74
Marksy Quick mark
Our fast mark
73
ChatGPT
GPT-6.1 Sol, criteria and essay pasted in
65
Claude
Claude Opus 5.5, criteria and essay pasted in
61
Gemini
Gemini 3.8 Flash, criteria and essay pasted in
57

0 = marks given at random. 100 = as close as a teacher.

Where each grader goes wrong

The share of essays each grader marked within one grade, split by the grade the IB gave. The chatbots hold up on middle grades and fall apart on 7s, the essays where a wrong mark costs most.

IB grade	3	4	5	6	7	All grades equally
Marksy Full review	72%	91%	91%	98%	65%	83%
Marksy Quick mark	63%	91%	96%	100%	68%	84%
ChatGPT	78%	100%	91%	82%	35%	78%
Claude	100%	98%	68%	73%	29%	74%
Gemini	56%	78%	89%	91%	74%	78%
A teacher	88%	90%	90%	100%	100%	94%

Essays per grade: 32 at 3, 58 at 4, 47 at 5, 45 at 6, 31 at 7. TOK essays and exhibitions are graded A to E, counted here as 5 to 1.

What “accurate” means here

Within one grade

The number most AI graders publish: 92% for Marksy. If the IB gave your IA a 5, Marksy's mark was 4, 5 or 6 for 92% of the 1,317 essays we tested. Because most IAs score a 5 or 6, we also check every grade on its own.

Every grade counted equally

The same question, but a 7 counts as much as a 5, so a grader can't do well by playing it safe. On all 1,317 essays Marksy scores 88%. On the 226 balanced essays it's 83%, against 78% for ChatGPT, 78% for Gemini and 74% for Claude.

Marking score

How far a grader's marks are from the IB's on average, with weak, middle and strong essays counted equally, turned into a score out of 100. 0 is what random marks get. 100 is as close as a student's own teacher. It's the measure we use to improve Marksy, because a grader can only score well by telling weak and strong essays apart.

Big misses

Out of 100 essays, how many a grader marks two or more grades away from the IB, with each grade from 3 to 7 counted equally. These are the marks that would send you in the wrong direction.

Why a teacher is so hard to beat

The IB doesn't re-mark every IA. Your teacher marks the whole class, the IB re-marks a small sample, and if the teacher was too generous or too harsh, every mark in the class moves by the same amount. So for most IAs, the IB's final mark is your teacher's mark, adjusted.

That means the answer key is partly made from the teacher's own marks, a head start no outside grader can have. Teachers also know the student, the task and every earlier draft. That's why a teacher is the 100 on our marking score: it's the ceiling, not a fair comparison.

What Marksy still gets wrong

It marks strong work a little low. 65% of grade 7 essays came back within one grade.
It marks weak work a little high. 72% of grade 3 essays came back within one grade.
Your teacher is still closer: they matched the IB's exact grade 68% of the time, Marksy 42%. The test favours teachers, because the IB's mark starts from theirs. Why

How we tested

1,317 IAs and TOK pieces that students shared publicly with the mark the IB gave them after moderation.
Marksy never saw the answer for an essay it was scored on: the settings used to mark each essay were fitted on other essays.
For the chatbots: the official criteria, the essay text and a request to mark it like an IB examiner. Default settings, one try each, run in November 2026.
Every grader saw the same text from the same files. Marksy used the same models and criteria it uses for you.

Tested

Biology, Chemistry and Physics IAs, Maths AA and AI explorations, Business Management IA, Economics IA, Geography IA, History IA, TOK essay, TOK exhibition.

Not tested yet

Extended Essays, Psychology IA (the essays we hold use the old criteria), Language and literature subjects.

Why we publish all of this

We want Marksy to be the most open IA grader there is. So this page shows the numbers that make us look good and the ones that don't, and says exactly how we tested.

Marksy doesn't send your draft to one chatbot and pass on its answer. Every grading goes through several steps, and each one has been tuned and tested, again and again, against essays with real IB marks. That work is why it marks closer to the IB than pasting your IA into ChatGPT, Claude or Gemini.

We keep going. Our research is ongoing, and every change is tested against the same essays before it reaches you, with most of the effort on the weak spots above: strong work marked too low and weak work marked too high. Some of the gap to a teacher can't be closed from the outside, but we close what we can, and this page changes when the numbers do.

How to check any AI grader's accuracy

Does it split results by grade? A grader that gives everyone a 5 or 6 looks fine until you look at the 3s and 7s.
Were the test essays kept out of training? Otherwise the grader has seen the answers.
Where do the marks come from? IB marks after moderation, a teacher's mark or the company's own reader are very different answer keys.
Which criteria were used? The Extended Essay criteria change for May 2027.

Marksy vs MyRevisionAgent

MyRevisionAgent says its IA grader lands within one grade 92% of the time. Marksy's is 92%. They don't publish results by grade, so you can't see how they do on weak or strong essays.

We sent both a polished Business Management EE where every theory, method and number was wrong. MyRevisionAgent gave it 22.1/34, a grade B. Marksy gave it 7/30.

See the full comparison

Questions

What does “within one grade” mean?: The IB turns your IA mark into a grade from 1 to 7 (A to E for TOK). If the IB gave your IA a 5, Marksy is within one grade when it says 4, 5 or 6. Marksy did this for 92% of the 1,317 essays we tested.
Is Marksy more accurate than ChatGPT?: In our test, yes. We pasted the official criteria and the same 226 essays into ChatGPT, Claude and Gemini. Their marks were 18% to 33% further from the IB's than Marksy's. ChatGPT and Claude also got far more grade 7 essays wrong.
Is Marksy as accurate as my teacher?: Not yet, and it isn't a fair race. For most IAs, the IB's final mark is your teacher's mark, moved up or down for the whole class after the IB re-marks a sample. So a teacher is checked against an answer that starts from their own mark. On our test, a teacher's mark matched the IB's exact grade 68% of the time, against 42% for Marksy. Marksy is useful because you can check a draft whenever you want, not because it replaces your teacher.
Is Marksy just ChatGPT with a prompt?: No. Your draft goes through several steps, and each one has been tuned and tested against essays with real IB marks. When we pasted the official criteria and the same essays into ChatGPT, its marks were 18% further from the IB's than Marksy's.
Does this cover the Extended Essay?: Not yet. The Extended Essays we hold are almost all high-scoring, so they can't show whether a grader tells weak and strong essays apart. We will add EE results when we have a fair set.
Where do the marks you tested against come from?: Students shared their IAs and TOK pieces publicly together with the mark the IB gave them after moderation. We use those marks as the answer key. Marksy never saw the answer for an essay it was scored on.
Why is the marking score out of 100?: It puts every grader on one scale that's easy to read. 0 is what you'd get from random marks and 100 is as close as a student's own teacher. Under the hood it's the average distance from the IB's mark, with weak, middle and strong essays counted equally.

Last tested November 2026. We re-run these tests when we change how Marksy marks, and update this page.

See where your draft stands

Upload your IA, EE or TOK essay and get a mark for every criterion, with comments on your own sentences.

Grade my draft