Papers With Code was, for years, one of the most useful free resources in machine learning. It linked research papers to their open-source code and maintained thousands of benchmark leaderboards tracking the state of the art across tasks, so a researcher could see at a glance which method currently led on a given benchmark and find the code to reproduce it. Founded by Robert Stojnic and Ross Taylor, it was later acquired and run by Meta.
In July 2025 Meta sunset the service. The site began returning errors and then redirected to Hugging Face’s Trending Papers feature; today paperswithcode.com no longer serves its old leaderboards and code links. The official data repository now points downloads to a Hugging Face organization, pwc-archive, which describes itself as a “public archive” containing “the last publicly available snapshot of the PWC datasets” preserved “before the original platform became unavailable.”
The shutdown drew a sharp reaction from researchers who relied on the leaderboards for tracking progress and reproducibility, and many felt the replacement focused on trending papers rather than the comprehensive, task-by-task state-of-the-art tracking that Papers With Code had provided. The episode left the community without a single canonical, free home for benchmark results.
Why business readers should care: Papers With Code shows how much shared research infrastructure can quietly depend on one company’s willingness to keep it running, and how its withdrawal can remove a public good that an entire field had built habits around.