Wouze · Open weights, local

laya-ara-rag

laya-ara-rag by Wouze scores 67.4 on the Arabic datasets, the only ones it ran. Calibration error 8.6%.

System One Index

–

Ran only some datasets

English

–

Arabic

67.4

Vision Index

–

Not run on images

Calibration error

8.6%

Lower is better

Median latency

58 ms

Local, not comparable

Cost per 1,000 calls

Local

As billed

Text datasets

Dataset	Accuracy	Rank	Best on the board	Calibration error
XNLI Arabic Arabic	70.0	15th of 26	89.3	8.7%
MASSIVE scenario Arabic	60.5	15th of 26	89.5	22.3%
MASSIVE intent Arabic	55.0	21st of 26	91.5	23.5%
OSACT4 offensive Arabic	84.5	15th of 26	95.5	5.0%
Tweet sentiment Arabic	44.0	24th of 26	80.0	17.0%
MIRACL rerank Arabic	71.5	18th of 26	90.5	11.3%
MIRACL pairwise Arabic	86.5	12th of 26	91.0	3.7%

Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.