LightOn Open-Sources Multilingual Retrieval Models Topping BEIR & MTEB Leaderboards
antoine_chaffin · x · 2026-07-30
LightOn has released mDenseOn and mLateOn, two fully open-source 307M-parameter multilingual retrieval models. These models achieve state-of-the-art results across English general-domain retrieval (BEIR), long-document retrieval (MLDR), multilingual retrieval (MIRACL), and code retrieval (MTEB Code).
Built upon a validated English data recipe, the pipeline was extended to eight additional languages via translate-train, creating a massive 2.8B-pair corpus. Most notably, the late-interaction model (mLateOn) demonstrates strong zero-shot generalization, remaining competitive in languages and scripts entirely absent from retrieval training (such as Cyrillic, Japanese, Korean, and Chinese). This effectively removes the main limitation of translate-train pipelines with dense models. Models, datasets, and training code are fully open-sourced.
Related event: LightOn Open-Sources Multilingual Retrieval Models mDenseOn and mLateOn(18 posts)→
More from Research
- SimpleMark: multi-bit LLM watermark bounds per-token distortion with f-divergence budget — rishabh16_ · 2026-10-16
- Stanford paper: generative representation learning cuts tokenizer training compute by 7-8x — HirokatuKataoka · 2026-10-16
- Cohere Labs to present confidence estimation for reasoning models at EMNLP 2026 — Cohere_Labs · 2026-10-16
- Vals RSI Index extrapolates Google models hitting 130% autonomous AI R&D by Oct 2027 — HaydnBelfield · 2026-10-16
- SpatialClaw Pulled Into Agentic Vision Data; Filtering and Evals Were the Real Lift — CMHungSteven · 2026-10-16
- Visual Lean: drag-and-drop GUI for the Natural Numbers Game runs Lean 4 in WASM — MilesCranmer · 2026-10-16