Most Liked AI Models on Hugging Face 2026

What this chart shows. The fifteen model repositories with the most likes on Hugging Face, read on 21 September 2026 at 08:19 UTC. Qwen3.8-27B leads with 15,900 likes, just ahead of the image model FLUX.1-dev on 14,862 and DeepSeek-R1 on 14,275. Five of the fifteen make images or video and seven write text, three of those reading images as well; the remaining three cover speech synthesis, speech recognition and search. The fifteen hold 128,232 likes between them.

How the data is collected. A like is left by a signed-in Hub account on a repository, in the way a star works on a code host, and the count is public. The Hub's Python library treats it as a relationship between a user and a repository: list_liked_repos returns the models, datasets and Spaces a given user has liked, and list_repo_likers returns the users who have liked a given repository. The chart uses the likes field of the public API with the model list sorted by likes, so the ranking is the Hub's own rather than anything computed here.

Background. Likes and downloads describe different things, and on this hub they disagree sharply. all-MiniLM-L6-v2, the most downloaded model on the whole platform at 251,048,129 downloads in thirty days, is only eleventh by likes with 6,095. In the other direction Meta's Meta-Llama-3-8B has 6,691 likes on 267,816 downloads in the same period, the smallest download count in this group. Five of the fifteen repositories were created during 2026.

Limits. Likes accumulate and are never reset, so a repository that has been on the Hub for years has had longer to gather them: nine of these fifteen were created in 2024 or earlier. A like counts an account, not a deployment, and nothing stops a team from publishing the same weights in several repositories, which splits the likes between them while each copy keeps its own downloads. The ranking also covers public repositories only, and the top of it is tight: 1,038 likes separate first place from second.