Model Publishers

Every model in the hub came from somebody, and this page sorts the catalog by who that somebody is. Across 1,692 publishers we hold 2,760 models, 859 datasets and 254 papers, spread over 32 tasks and 18 licenses. Eight names sit at the top: Qwen with 91 models, AI at Meta with 71, Google and Convergent Intelligence with 64 each, Microsoft with 47, PyTorch Image Models and NVIDIA with 42 apiece, and the Helsinki-NLP Research Group with 35. We keep this view because the publisher decides two things an operator cannot change after the download: the license and what is in the files. Plan a rack around a model family and you are committing to the group that ships it.

Start with the task you run, not the name you recognize. Text generation leads at 338 models, image and text to text at 138, robotics at 81 and speech recognition at 78. Pick the task, open a publisher, and weigh its models against our four collections. NVIDIA and AI at Meta also publish datasets, six and five, which matters if you plan to fine-tune in your own facility. GPU prices were read on Sep 18, 2026.