Global PIQA: Results as Published

Global PIQA is a participatory commonsense benchmark in the PIQA format: a prompt and two candidate solutions, one correct and one incorrect, that need physical, temporal, cultural, or everyday knowledge. Researchers who speak the languages wrote a non-parallel split for 136 language varieties, with little or no translation, and a parallel split translates more culturally agnostic questions into 131 varieties. The full set covers 141 language varieties, and native speakers checked the examples. The score is accuracy, reported as an average across languages, either from completion probabilities or from a prompted answer.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.