Scores cover the same development + test release. Hard scores use the tagged subset; small task counts limit precision. Rankings do not establish a Greek-proficiency pass threshold.
Dataset revision: 66e5f4ac06d5c2f50eb42f7cd78c8af1af4468e2
Hard-track figures are development calibration results. Independent proficiency validation remains outstanding.
Adversarial development calibration on a limited task subset. The macro averages only represented NLU tasks; check Tasks and Items before comparing scores. These are not independent proficiency results.
© KIEFER TEK LTD · numbers only, no benchmark items are published · data: KIEFERSA/greek-nlu-bench