AIME GEPA: Algorithmic + Path D — 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 0.3533

pass@2 0.4533

pass@3 0.5067

pass@4 0.5267

pass@5 0.5267

pass@6 0.5600

pass@7 0.5800

pass@8 0.5933

pass@9 0.5933

pass@10 0.6000… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-degeneration-dev/aime-2025-gepa-prompta-plus-pathd.