Skip to content

Benchmark

LuxIT base WNLI F1-macro

Multilingual benchmarks compare model performance across languages. Higher scores are better.

Models and systems measured
14
Scores recorded
14
Sources
1
IQ weight
No weight published

Score distribution

0.379
Mean
0.331
Median
0.103
Std dev
0.265
10th percentile
0.52
90th percentile
0.265
Min
0.2650.569
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Apertus-8B-Instruct-2509table: 10, column: WNLI, system: apertus-8b· LuxIT (arXiv:2510.24434v4) data0.265
EuroLLM-1.7B-Instructtable: 10, column: WNLI, system: eurollm-1.7b· LuxIT (arXiv:2510.24434v4) data0.318
Gemma-3-12B-ITtable: 10, column: WNLI, system: gemma-3-12b· LuxIT (arXiv:2510.24434v4) data0.569
Gemma-3-1B-ITtable: 10, column: WNLI, system: gemma-3-1b· LuxIT (arXiv:2510.24434v4) data0.293
GLM-4-9B-0414table: 10, column: WNLI, system: glm-4-9b· LuxIT (arXiv:2510.24434v4) data0.478
Llama-3.1-8B-Instructtable: 10, column: WNLI, system: llama-3.1-8b· LuxIT (arXiv:2510.24434v4) data0.52
Llama-3.2-1B-Instructtable: 10, column: WNLI, system: llama-3.2-1b· LuxIT (arXiv:2510.24434v4) data0.459
Ministral-3-3B-Instruct-2512table: 10, column: WNLI, system: ministral-3-3b· LuxIT (arXiv:2510.24434v4) data0.329
Mistral-7B-Instruct-v0.3table: 10, column: WNLI, system: mistral-7b· LuxIT (arXiv:2510.24434v4) data0.265
Olmo-3-7B-Instructtable: 10, column: WNLI, system: olmo-3-7b· LuxIT (arXiv:2510.24434v4) data0.285
Phi-4table: 10, column: WNLI, system: phi-4· LuxIT (arXiv:2510.24434v4) data0.293
Qwen2.5-0.5B-Instructtable: 10, column: WNLI, system: qwen2.5-0.5b· LuxIT (arXiv:2510.24434v4) data0.331
Qwen2.5-1.5B-Instructtable: 10, column: WNLI, system: qwen2.5-1.5b· LuxIT (arXiv:2510.24434v4) data0.385
Qwen2.5-7B-Instructtable: 10, column: WNLI, system: qwen2.5-7b· LuxIT (arXiv:2510.24434v4) data0.511

Data credits: Julian Valline, Cedric Lothritz, Siwen Guo and Jordi Cabot. LuxIT: A Luxembourgish Instruction Tuning Dataset from Monolingual Seed Data. arXiv:2510.24434v4. CC BY 4.0 article numeric/protocol extract; restructured, printed values unchanged. Task data, private fine-tunes, exams and weights excluded. (licence)