AIME GEPA: Algorithmic RLM (A/B/C Paths) — AIME 2025 prompt evaluation results from GEPA prompt optimization. Evaluated on 150 AIME 2025 problems (MathArena/aime_2025 x5) with 10 samples per problem using openai/gpt-4.1-mini at temperature 0.7.

	Results

Metric Value

pass@1 0.3667

pass@2 0.4467

pass@3 0.4667

pass@4 0.5000

pass@5 0.5400

pass@6 0.5600

pass@7 0.5733

pass@8 0.5867

pass@9 0.5867

pass@10 0.6067… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/aime-2025-gepa-prompt-a-iter2.