{
  "id": 2153943,
  "title": "Two Million Open AI Models, but Most of the Attention Goes to Just 200",
  "url": "https://urgent.news/2026/08/20/two-million-open-ai-models-but-most-of-the-attention-goes-to-just-200",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-20T14:24:37.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/ihopkins/two-million-open-ai-models-but-most-of-the-attention-goes-to-just-200-141j"
  },
  "original_language": "en",
  "account": "Hugging Face has reported impressive growth, boasting over 13 million users, more than two million public models, and more than 500,000 datasets in 2025. However, the actual usage patterns reveal a stark contrast. Approximately half of the models on the platform have garnered fewer than 200 downloads, while just 0.01% of models - roughly 200 - account for a staggering 49.6% of all downloads. This indicates a highly concentrated ecosystem despite the vast number of available models.\n\nThe proliferation of models may create a perception of abundant choices, but in reality, the usable ecosystem is substantially smaller. Developers tend to gravitate towards well-performing, well-documented models with robust support, fostering valuable network effects. However, this does not necessarily mean that the smaller group of highly downloaded models is inherently superior.\n\nWhen evaluating open or open-weight models, it is crucial to consider more than just benchmark scores and parameter count. Factors such as recent maintenance, compatibility with established inference frameworks, available quantisations, clear licensing, and community-created fine-tunes or adapters become essential indicators of a model's production readiness. These signals demonstrate whether a model is part of an actively improving ecosystem or an isolated repository.\n\nWhile download volume can be a useful signal, it should not be the sole determinant in selecting a model. A highly specialised model may have modest downloads but still be highly valuable for its intended users. Conversely, a widely downloaded model may not be the optimal choice for every workload, as popularity does not guarantee suitability for specific latency targets, GPU availability, or prompt reliability.\n\nThe concentration of downloads also raises concerns about supply-chain management. Thousands of applications may depend on a small number of base models, maintainers, and supporting tools. Should any of these crucial components change, become inactive, or suffer vulnerabilities, the impact could ripple through numerous applications, highlighting the importance of understanding a model's origin, maintenance, and potential migration options.\n\nUltimately, the open AI ecosystem faces a shortage of dependable ways to determine which models deserve long-term trust. The sheer number of published models does not equate to unlimited choice. The next phase of the ecosystem will depend on developers' ability to evaluate, maintain, secure, and deploy models effectively, ensuring the presence of sufficient evidence, tooling, and community support to guarantee reliability in production environments.",
  "summary": "Every time I open Hugging Face, the model ecosystem seems to have expanded again. There are new base models, fine-tunes, quantisations, adapters and experimental releases arriving constantly. At first glance, this looks like an incredibly diverse market. We have more models, more developers and more ways to run AI than ever before. But when I looked at where the downloads actually go, I found a…",
  "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."
}