Commonsense Reasoning research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

CoS-E v1.0

T5-base

0.695Score (Finetune+Baseline vs Predict+Baseline)

Feb 26, 2026

CoS-E v1.11

T5-base

0.608Score (Finetune+Baseline vs Predict+Baseline)

Feb 26, 2026

StrategyQA

zero-shot-cot

54.8Accuracy (format-specific prompt)

Feb 27, 2026

Winogrande

Flexora

3.84Time (h)

Feb 26, 2026

HellaSwag, WinoGrande, CommonsenseQA, OpenBookQA, PIQA, SIQA

—

—Primary metric

Apr 29, 2026

HellaSwag, WinoGrande, BoolQ, CommonsenseQA, OpenBookQA, PIQA, SIQA

—

—Primary metric

Apr 29, 2026

HellaSwag 77 (test)

—

—Score

Feb 26, 2026

XStoryClozee 75 (test)

—

—Score

Feb 26, 2026

XCOPA 74 (test)

—

—Score

Feb 26, 2026

HellaSwag Māori

—

—Accuracy

Feb 26, 2026

XStoryCloze Māori

—

—Accuracy

Feb 26, 2026

CSQA v1.1 (test)

—

—Accuracy

Feb 26, 2026