pmahdavi/

aime2025-merging-leaderboard

Evaluation results for 8 models on AIME 2025 with 30 problems and 32 rollouts per problem.

We followed the evaluation guidelines and prompts from OLMo 3. Best effort was made to ensure reported numbers are as accurate as possible. Code: pmahdavi/modal-eval

	Leaderboard




	

Model pass@32 avg@32 Loop Rate

allenai/Olmo-3.1-7B-RL-Zero-Math 80.0% 38.0% 2.9%

pmahdavi/Olmo-3.1-7B-Math-Code 73.3% 36.4% 2.4%… See the full description on the dataset page: https://huggingface.co/datasets/pmahdavi/aime2025-merging-leaderboard.