AIME GEPA: Inputs-Only RLM Final — 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 0.3400

pass@2 0.4200

pass@3 0.4733

pass@4 0.5333

pass@5 0.5400

pass@6 0.5533

pass@7 0.5533

pass@8 0.5667

pass@9 0.5667

pass@10 0.5800… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/aime-2025-gepa-inputs-only-rlm-final.