Leaderboard

SomBench model rankings

How open multilingual and Somali-specific models perform across six core Somali NLP tasks.

Preview benchmarks. Scores below are from internal SomBench v0.1 runs. Public model weights and reproducible eval scripts are releasing soon.
#ModelParamsOverall score
1
goobo-baseSomali-native
Goobo Labs
125M
52.4
2
afroXLMR-large
Masakhane
561M
44.1
3
mDeBERTa-v3-base
Microsoft
86M
38.7
4
XLM-RoBERTa-base
Meta AI
125M
35.1
5
mBERT
Google
110M
32.4
Scores are internal SomBench v0.1 preview averages on a held-out test set. Higher is better. Submit your model via the contact form.
Build with us

Build Somali AI in the open

Use our datasets and models, contribute to the research, or partner with the lab. Everything we can open, we do.