tyrtleli/

thinking-benchmark-gpt5-2

Five problems selected for the k=32 deep pass of the Cost of Overthinking study, targeting GPT-5.2's capability edge (25–75% success rate on the k=8 calibration pass). Subset of tyrtleli/thinking-benchmark-90.

	Problems




	

id source domain difficulty k=8 rate answer

olymmath_0552 OlymMATH algebra 5 2/8 \frac{10057600}{3}

hmmt_2026_feb_comb_08 HMMT combinatorics 5 2/8 6992

olymmath_0518 OlymMATH number_theory 4… See the full description on the dataset page: https://huggingface.co/datasets/tyrtleli/thinking-benchmark-gpt5-2.