Hugging Face revives `PapersWithCode` with AI-parsed leaderboards
The rebuilt site tracks trending papers, methods, citations, repos, artifacts, and benchmark results. Useful for model scouting, but still manually verified and early-stage.
- Default ranking uses GitHub star velocity, so it surfaces research projects gaining developer attention, not just citation-heavy papers.
- Coverage starts with high-impact items like
Qwen 3.5,RF-DETR,DINOv3,MTEB,Open ASR Leaderboard, and coding-agent benchmarks. - Paper pages auto-link GitHub repos, project URLs, artifacts, PDFs, and external non-Arxiv papers; multiple repos per paper are supported.
- Leaderboards exist by benchmark and domain, including
MMTEB,COCO val 2017, andTerminal Bench; handy for fast model/vendor filtering. - Result extraction uses AI agents, but verification is still manual. Treat it as a shortlist generator, not a source of record yet.