SkMTEB gives Slovak its first MTEB-style embedding benchmark, and a reusable recipe for low-resource languages.

SkMTEB (arXiv 2606.13647, Šuppa et al.) is the first comprehensive embedding benchmark for Slovak, paired with model-adaptation experiments. Beyond the specific language, it's a template for evaluating and adapting embedding models when you don't have English-scale data. If you're doing non-English RAG or multilingual retrieval, the methodology transfers even if the language doesn't.