Model

DeepSeek-Coder-Instruct 6.7B

DeepSeek · open weights

Knowledge & Reasoning

Benchmark	Score	Metric	Harness	Source	Confidence	Link
MMLU	36.4	exact_match	N/A	Independent	medium	arxiv.org
WinoGrande	57.6	accuracy	N/A	Independent	medium	arxiv.org

Benchmark	Score	Metric	Harness	Source	Confidence	Link
GSM8K	21.3	exact_match	N/A	Independent	medium	arxiv.org

Benchmark	Score	Metric	Harness	Source	Confidence	Link
APPS	33.8	introductory_pass@1	N/A	Independent	medium	arxiv.org
MBPP	65.4	accuracy	N/A	Independent	medium	arxiv.org
SAFIM	63.4	average	N/A	Independent	medium	arxiv.org