Changelog

Benchmark datasets can now host evaluation leaderboards, and models can surface their evaluation scores right on the Hub. You can submit PRs with results from evaluations you have conducted to make them visible as community evals.

Check out the docs to learn more.

You can now access all your blog drafts directly from the editor, neatly grouped by the organizations you’re part of or under your personal account. This makes it easier to manage, edit, and pick up where you left off.

You can now see the number of running jobs directly from the user menu, giving you quick visibility into your active workloads without leaving the current page. This makes it easier to monitor activity and understand which jobs are currently running.

Datasets can now be sorted by "Largest total size" and "Smallest total size" when searching for datasets or from users and organizations profile pages. It uses the sum of the repository large file size, on the main branch.

Model pages now display hardware compatibility for MLX models. Logged-in users see compatibility indicators based on their configured hardware, with clear warnings when Apple Silicon is required.

This brings MLX in line with GGUF, helping you quickly identify which variants will run on your machine before downloading.