AI Grading Benchmark
Learning Design Alliance · benchmark

Can an AI grade student work, and how does it fail?

We gave 22 AI models the same real student work (short written answers, computer code and essays) and asked each to grade it with the same simple instructions. Then we checked two things: do the AI's grades match the grades experienced teachers gave, and can the AI be fooled, for example by a note in the answer asking for full credit, by a blank answer, or by a student who simply argues?

As of October 2026. These findings describe the model versions tested in October 2026 (each model's page names the version). Findings about AI models can go out of date quickly: when 50 widely cited findings were re-tested on newer models, 44% no longer held, typically about 16 months after they were published . So we plan to re-test each finding here as new models are released, rather than treat it as settled.

What these results show, and what they do not They show how a setup, one model with one prompt and its settings, grades these cases. They do not show that AI grading helps students learn; that a model "grades as well as teachers" (agreement near the raters' own means it disagrees with them about as often as they disagree with each other, on these cases); how it will do on another subject, age group, language or prompt; or that it is fair to any group of students. A check that finds no effect states the largest effect it cannot rule out. Please cite a strand and a setup, not the overall score alone.
Three ways a result can vary Asking twice. An AI model can give the same answer two different grades. We graded every answer twice, independently; the Consistent score and the Same grade twice column show how often the two grades matched. A different set of answers. Each task uses a sample of about 100 student answers. A different sample would shift every score a little. The ranges shown beside results (and Could rank on the leaderboard) show how far; when two models' ranges overlap, we cannot tell which is better. Different wording. Every model got the same instructions. Rewording them changes some models' grades more than asking twice does; see What changes results.

Top of the board

Ranked by the overall score. A model that failed a basic safety check (it could be talked into higher grades, or it gave credit to blank answers) is placed below every model that passed. See the full leaderboard.

Every number here can be checked: each model's grades are saved, and the scores are recalculated from them. Before looking at any real result, we tested every check on simulated graders with a known flaw built in, to make sure the check could catch it.