Model

GPT-NeoX 20B

OpenAI · closed weights

Knowledge & Reasoning

Benchmark	Score	Metric	Harness	Source	Confidence	Link
ARC-Challenge	45.4	accuracy	N/A	Independent	medium	arxiv.org
ARC-Easy	70.8	accuracy	N/A	Independent	medium	arxiv.org
BoolQ	64.9	score	N/A	Independent	medium	arxiv.org
BoolQ	46.4	accuracy	N/A	Independent	medium	arxiv.org
commonsense_qa	60.4	accuracy	N/A	Independent	medium	arxiv.org
HellaSwag	70.5	overall_accuracy	N/A	Independent	medium	arxiv.org
HellaSwag	68.4	accuracy	N/A	Independent	medium	arxiv.org
MMLU	33.6	average_(%)	N/A	Independent	medium	arxiv.org
MMLU	33.6	exact_match	N/A	Independent	medium	arxiv.org
MultiRC	22.9	f1	N/A	Independent	medium	arxiv.org
OpenBookQA	38.8	accuracy	N/A	Independent	medium	arxiv.org
PIQA	76.7	score	N/A	Independent	medium	arxiv.org
PIQA	75.8	accuracy	N/A	Independent	medium	arxiv.org
RTE	53.8	accuracy	N/A	Independent	medium	arxiv.org
WinoGrande	66.1	accuracy	N/A	Independent	medium	arxiv.org