blink-multilingual-64M scores 51.60 on MMTEB, beating potion-128M with half the parameters
tomaarsen · x · 2026-11-18
tomaarsen cites MinishLab's evals: on MMTEB's 131 tasks, blink-multilingual-64M averages 51.60 vs 47.67 for potion-multilingual-128M (both 256d) — a higher score with half the parameters, and it still leads after removing the 35 English-only tasks.
model2vec, MinishLab's open-source library (2.3k GitHub stars, MIT), distills any Sentence Transformer into a static embedding model — up to 50x smaller and 500x faster with minimal quality loss.
Related event: MinishLab Releases Blink Static Embeddings for Fast CPU Retrieval(5 posts)→
More from Models
- Every's founder pushes back: we re-tested the model to the wire, OpenAI gave us no heads-up — kieranklaassen · 2026-11-18
- HF trending: Thinking Machines' Inkling-2-Small, MiniMax H3-Image and Qwen3.8-27B surge together — victormustar · 2026-11-18
- After DeepSeek's rate limits: a 4-step guide to the official paid desktop path — sujingshen · 2026-11-18
- blink-8M tops seven static embedding models with 55.35 MTEB and 46.70 BEIR nDCG@10 — tomaarsen · 2026-11-18
- blink-multilingual-64M demo: English query retrieves Dutch and Spanish docs correctly — tomaarsen · 2026-11-18
- MinishLab ships Blink static embedding models: 8M and 64M params, fast CPU retrieval — tomaarsen · 2026-11-18