MTEB, but for style: STEB, the Style Text Embedding Benchmark #EMNLP2026. 96 datasets, 7 languages, 40 models. Good at semantics ≠ good at style. A top-5 MTEB model ranks last. arxiv.org/abs/2606.31741
AI Weekly's analysis
→
- STEB bundles 96 datasets across 7 languages to standardize how style text embeddings are evaluated.
- The paper finds semantic embeddings 'consistently fail' at stylistic tasks like authorship verification and AI-text detection.
- No single style embedding was universally best across STEB's tasks, which include authorship retrieval and linguistic-feature probing.
Read full analysis →
