GSM-Hard

Dataset Summary This is the harder version of gsm8k math reasoning dataset (https://huggingface.co/datasets/gsm8k). We construct this dataset by replacing the numbers in the questions of GSM8K with larger numbers that are less common.
Supported Tasks and Leaderboards This dataset is used to evaluate math reasoning La

Papers using GSM-Hard (2)

The Cost Of Consensus: Isolated Self-correction Prevails Over Unguided Homogeneous Multi-agent Debate2026Compiling Deterministic Structure Into SLM Harnesses2026