{
  "id": 19679407,
  "title": "A 23M-Parameter Model Still Wins: 338 Embedding Models Ranked by Real Downloads",
  "url": "https://urgent.news/2026/11/12/a-23m-parameter-model-still-wins-338-embedding-models-ranked-by-real",
  "topic": "culture",
  "section": "Culture",
  "published": "2026-11-12T16:50:13.000Z",
  "source": {
    "name": "HackerNoon",
    "slug": "hackernoon",
    "url": "https://hackernoon.com/a-23m-parameter-model-still-wins-338-embedding-models-ranked-by-real-downloads?source=rss"
  },
  "original_language": "en",
  "account": "all-MiniLM-L6-v2, a 23 million parameter model from sentence-transformers, garnered 177 million downloads in the past 30 days, comprising 38% of all downloads across the 213 open-weight embedding models monitored by the TextModels directory. It outpaced the other four most downloaded models combined. Below is a list of the 338 models in the directory, ranked by downloads, followed by hosted APIs, Ollama builds, and historical models. While benchmark scores are often used to rank the \"best\" embedding models, download counts reveal what developers actually use in production. The sentence-transformers series has historically led the way, with their descendants continuing to dominate download charts. Small and older models often outperform big and newer ones in terms of downloads. The top two downloaders, all-MiniLM-L6-v2 and BAAI's bge-small-en-v1.5, are both quite small. Most production searches prioritize speed and adequacy over raw performance. The market is heavily concentrated, with the top 10 open-weight embedders accounting for 74% of the 470 million monthly downloads in this list. A newcomer to watch is Qwen3-Embedding-0.6B, which already ranks ninth with 9 million monthly downloads and its 4B and 8B siblings each surpass 2.4 million downloads. Multilingual models are mainstream, with five of the top 12 being multilingual. Various vector databases incorporate their own model builds, showing how many teams rely on databases for embeddings rather than selecting a model directly. This list was compiled from the TextModels directory's Embedding filter as of November 11, 2026. Hugging Face entries are ranked by 30-day downloads obtained from the Hugging Face API. Models with verified metrics are marked as \"Verified.\" Hosted API models lack public download counts, so they are grouped by provider alphabetically. Some hosted entries, like OpenAI's 001-series search models, have been retired but remain in the list due to continued relevance when migrating old pipelines.",
  "summary": "all-MiniLM-L6-v2 got 177M Hugging Face downloads in 30 days. See all 338 embedding models ranked by real downloads, from BAAI and Qwen to OpenAI.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}