ML Engineering & Research
MLRC-Bench
7 ML research-competition tasks testing whether language agents invent novel methods.
Response matrix
Every model, scored item by item.
Each row is an AI model and each column an item, ordered so the strongest models and easiest items gather toward one corner. 7 subjects × 7 items, 100% of cells evaluated.
Fit to width. Hover for subject & item; click a cell for details.
Scale: Improvement as a fraction (can be negative or > 1): absolute improvement over the baseline, or improvement relative to the human top solution — each on its own scale.
Sample items
What the questions look like — and how subjects answer.
A spread of items across the difficulty range. This benchmark does not publish per-answer traces, so each item shows which subjects succeeded.
Subjects
The models, agents, and reward models evaluated.
7 subjects, ranked by mean response across this benchmark's items.
- 16bc8c1a40.171
- 2c2a34b6e0.134
- 31b2233840.13
- 45967999d0.123
- 5d63974f50.108
- 67505ef580.078
- 7159622a40.075