179M-Parameter Open Retriever UIBK-Embedding-v1 Tops BEIR Under 1B Models
antoine_chaffin · x · 2026-10-13
Researchers at the University of Innsbruck released UIBK-Embedding-v1, an open English retrieval model (179M params, Apache 2.0) that packs both a dense retriever and a ColBERT late-interaction retriever into one encoder, fusing their scores inside a single standard MaxSim search — one model, one index, one search, replacing the usual two-index hybrid setup.
- BEIR: 59.79 nDCG@10, the best among sub-1B models with complete official MTEB results; beats LateOn on all 15 datasets, including the decontaminated BEIR
- NanoBEIR: 70.55, #1 among models up to 6B parameters
- Loads as a normal PyLate ColBERT model and runs in a PLAID index with no changes to the index or scorer
More from Models
- Meta's Muse Wonder scores 21 on ExploitGym vs Claude models' 191-300 — scaling01 · 2026-10-13
- Researcher flags Google Astra's bloated explanations, floats a conciseness benchmark — OfirPress · 2026-10-13
- OpenAI's latest blow to mathematicians: 'They have annihilated my field' — ThereWas · 2026-10-13
- Google AI Overviews denied a person had died, renewing criticism of AIO for news — lilyraynyc · 2026-10-13
- User feeds Claude Opus GTA frame-by-frame screenshots since it can't watch videos — adonis_singh · 2026-10-13
- Gemini Deep Research on web is really slow today, user reports — gaganghotra_ · 2026-10-13