Papers with Code - AI research platform, automatically linking papers, code, and more

What is Papers with Code

Papers with Code isan AI research platform integrating papers, code, datasets, and leaderboards, whose core mission is to deeply link academic papers with open-source code and SOTA performance leaderboards. The platform aggregates arXiv papers, automatically links GitHub repositories and Hugging Face models, provides SOTA leaderboards categorized by task, and visualizes technological evolution. The Hugging Face team rebuilt it from scratch in May 2026, with AI agents automatically parsing papers and maintaining data, covering traditional CV/NLP as well as emerging fields such as Agent, making it a one-stop infrastructure for researchers to track the frontier and reproduce results.

Main features of Papers with Code

• Paper aggregation and search: collects papers from arXiv and other sources, supports filtering by field, task, and conference.
• SOTA leaderboards: compares model performance by task and dataset, and displays historical evolution curves.
• Code links: automatically links GitHub repositories, Hugging Face models, and datasets.
• Conference pages: such as the CVPR 2026 page, which aggregates all accepted papers (including Oral/Spotlight).
• AI agent-driven: automatically tags papers, extracts evaluation results, and updates leaderboards.

How to use Papers with Code

• Visit the official website: visit the Papers with Code official website https://paperswithcode.co/，进入首页。
• Browse trending papers: click Trending to view the popular research developments currently attracting community attention.
• Find SOTA leaderboards: go to Browse state-of-the-art and search model performance rankings by task or dataset.
• Filter conference papers: click Conferences to select conferences such as CVPR, and view all accepted papers and their Oral/Spotlight classifications.
• Get code and models: click the GitHub or Hugging Face links on the paper detail page to directly obtain open-source code and pretrained models.
• Track technological evolution: view the SOTA progression curves on the leaderboard page to understand the course of technological development in the field.
• Submit feedback: when you find data errors or omissions, click Submit feedback to help the community improve the information.

Core advantages of Papers with Code

• Automatically maintained by AI agents: deploys agents to automatically parse thousands of papers every day, extract tags, performance data, and code links, without the need for manual entry-by-entry maintenance, keeping pace with the speed of paper publication in real time
• Deep paper-code-benchmark linkage: integrates papers, open-source code, datasets, and SOTA leaderboards on one page, compressing the time for researchers to find reference implementations from days to minutes
• SOTA leaderboard visualization: provides multidimensional comparisons by task, dataset, and evaluation metric, and displays technological evolution curves to intuitively track field development
• Comprehensive coverage of emerging fields: in addition to traditional CV/NLP, it adds frontier benchmarks such as Agent and Coding Agent, keeping up with research hotspots
• Conference page aggregation: view all accepted papers from top conferences such as CVPR in one click, including Oral/Spotlight classifications, to quickly locate key results

Comparison of Papers with Code with similar competitors

Use Cases of Papers with Code

• Rapid Literature Research: Search for the latest papers by task or field, and obtain code and models with one click, greatly shortening the reproduction cycle.
• Model Selection Comparison: Horizontally compare model performance by dataset and metric through SOTA leaderboards to assist technical selection decisions.
• Tracking Technological Evolution: View SOTA progression curves to understand the development trends of the state of the art on specific benchmarks over the years.
• Conference Paper Overview: During top conferences such as CVPR, quickly filter key Oral and Spotlight papers through the conference page.
• Open Source Ecosystem Discovery: Automatically links Hugging Face models and datasets, making it easy to download them for trial use and secondary development.