115M MoE ColBERT beats Qwen3-Embedding-0.6B on 9 of 10 quality benchmarks
tomaarsen · x · 2026-10-26
VAGOsolutions released the SauerkrautLM-MoEdernColBERT family on Hugging Face — multi-vector ColBERT retrieval models built on a ModernBERT hybrid MoE architecture, covering German, English, Spanish, Dutch, French, Italian and Portuguese.
Per tomaarsen, the 115M model beats Qwen3-Embedding-0.6B on 9 of 10 quality rows, with English NanoBEIR the sole exception (0.643 vs 0.648), while the six-language average flips in its favor (0.585 vs 0.574).
All three sizes work with Sentence Transformers' MultiVectorEncoder via trustremotecode=True; the 1.5M model correctly matches German mountain and English capital queries with higher MaxSim scores indicating better matches.
Related event: VAGOsolutions releases tiny 1.5M-115M MoE ColBERT retrieval models(10 posts)→
More from Models
- Anthropic Confirms Claude Bypassed a Paywall to Access Georgia Voter Counts — TechNadu · 2026-10-26
- Gemma 4 26B-A4B runs offline on a 16 GB Mac: 33/40 on TypeScript tasks at 31 tok/s — Mission-Sir-1614 · 2026-10-26
- Claude safety classifier blocks SS7 attack analysis, dev pivots to abliterated GLM — l4rz · 2026-10-26
- Sakana AI to pretrain first domestic model Sakana Sawara for defense and cybersecurity by 2027 — tkasasagi · 2026-10-26
- 31M-param English TTS model phonon-31m-en trends on Hugging Face — gradium · 2026-10-26
- NVIDIA tutorial: post-training Nemotron for Europe with Persona datasets and LoRA — NVIDIA Developer · 2026-10-26