NLG Meta-evaluation research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

WMT en-is 24

Llama 3.3 70B

0.36Kendall Correlation

Mar 11, 2026

WMT en-cs 24

Llama 4 Scout

0.318Kendall Correlation

Mar 11, 2026

WMT 24 cs-uk

Llama 4 Scout

0.825Kendall Correlation

Mar 11, 2026

WMT xh-zu 21

Llama 3.3 70B

0.323Kendall Correlation

Mar 11, 2026

WMT en-ha 21

Llama 3.3 70B

0.386Kendall Correlation

Mar 11, 2026

RoSE

Qwen 3 30B

0.635Kendall Correlation

Mar 11, 2026

MOCHA

Qwen 3 30B

0.72Kendall Correlation

Mar 11, 2026

CUS-QA orig. (uk)

Qwen 3 30B

0.681Kendall Correlation

Mar 11, 2026

CUS-QA orig. (sk)

Qwen 3 30B

0.788Kendall Correlation

Mar 11, 2026

CUS-QA orig. (cs)

Qwen 3 30B

0.804Kendall Correlation

Mar 11, 2026

CUS-QA en uk

Qwen 3 30B

0.577Kendall Correlation

Mar 11, 2026

CUS-QA en (sk)

Llama 4 Scout

0.661Kendall Correlation

Mar 11, 2026

CUS-QA en cs

Llama 4 Scout

0.73Kendall Correlation

Mar 11, 2026