Machine learning website, papers, code, and benchmarks

Papers With Code

Papers With Code is a machine learning website and data project known for linking research papers to code implementations, datasets, methods, evaluation tables, benchmarks, and reproducibility resources.

Official domain

paperswithcode.com is the public domain associated with Papers With Code; on June 1, 2026, it redirected to Hugging Face Papers.

Core idea

The service made machine learning papers easier to inspect by connecting papers with implementations, datasets, methods, and evaluation tables.

Open data

The paperswithcode-data GitHub repository describes itself as the full dataset behind paperswithcode.com and links to data dumps hosted on Hugging Face.

Papers With Code is associated with machine learning papers, code implementations, datasets, methods, and benchmark metadata.View Papers With Code on GitHub

What Papers With Code is

Papers With Code is a machine learning website and research-data project associated with paperswithcode.com. It became known for connecting academic papers to code repositories, datasets, methods, evaluation results, and benchmark leaderboards so readers could move from a claim in a paper to an implementation or comparison table.

Current domain behavior

On June 1, 2026, requests to paperswithcode.com returned a redirect to Hugging Face Papers. That means the public domain no longer behaves like a standalone Papers With Code browsing site in the way many older links and references describe. The Papers With Code GitHub organization and data repositories remain important references for the project's dataset and tooling.

Papers, code, and reproducibility

The site's central habit was simple but powerful: put a paper near the code that tries to implement it. For machine learning, this matters because a paper's method, hyperparameters, datasets, preprocessing choices, and evaluation settings often decide whether a result can be reproduced or compared fairly.

Datasets and evaluation tables

Papers With Code organized tasks around datasets and evaluation tables. A reader could look up a task, inspect the datasets used for that task, and compare reported results across methods. Those tables were useful entry points, but they still required caution because benchmark scores can depend on splits, metrics, training data, model size, and reporting conventions.

Open data archive

The paperswithcode-data repository says it provides the full dataset behind paperswithcode.com. Its README links to data dumps for papers with abstracts, links between papers and code, evaluation tables, methods, and datasets, and states that the data is licensed under CC BY-SA 4.0.

APIs and tools

The Papers With Code GitHub organization includes tooling beyond the public website, including an API client, data extraction projects, model-index tools, and guides for releasing research code. These projects show how the website's value extended into machine-readable metadata and community workflows, not only human browsing.

Strengths and limits

Papers With Code was useful because it compressed a messy research workflow into links, tables, and task pages. Its limits are important: linked code may be unofficial, incomplete, stale, or hard to run; benchmark tables can mix subtly different settings; and a high score on one leaderboard is not the same thing as a robust or generally useful model.

Why it matters

Machine learning research moves quickly, and papers without code can be difficult to verify or build on. Papers With Code matters because it helped normalize the expectation that results should be connected to implementations, datasets, and evaluation details. Even when the domain experience changes, that expectation remains part of how many researchers read modern AI papers.

WHOIS domain data

Data pulled: June 1, 2026View current WHOIS record

Domain: paperswithcode.com
IP address: 13.226.238.73
Registrar: Amazon Registrar, Inc.
WHOIS server: whois.registrar.amazon
Referral URL: http://registrar.amazon.com
Created: June 15, 2018
Updated: May 16, 2026
Expires: June 15, 2027
Nameservers: ns-1354.awsdns-41.org (205.251.197.74); ns-1539.awsdns-00.co.uk (205.251.198.3); ns-357.awsdns-44.com (205.251.193.101); ns-891.awsdns-47.net (205.251.195.123)
Domain status: clientTransferProhibited
DNSSEC: unsigned
Contact privacy: Registrant, admin, technical, and billing contacts are listed through whoisproxy.com.

Key concepts

Paper-code linkA connection between a research paper and an implementation repository or project page.
BenchmarkA dataset, task, metric, and comparison setting used to evaluate methods.
Evaluation tableA structured table that compares reported results across models or methods.
MethodA named machine learning approach, architecture, algorithm, or technique tracked across papers.

Common workflows

Find whether a machine learning paper has linked implementation code.
Compare models on a benchmark task before reading individual papers in depth.
Download public metadata dumps for papers, code links, methods, datasets, and evaluation tables.
Use repository tools or API clients to build research dashboards, searches, or reproducibility checks.

Common misconceptions

A linked code repository does not guarantee that the paper is fully reproducible.
A leaderboard score is not a universal ranking of model quality.
Official and unofficial implementations should be treated differently when interpreting results.
The current redirect from paperswithcode.com to Hugging Face Papers does not erase the historical Papers With Code dataset and tooling.

Open questions

How should benchmark sites show uncertainty, compute cost, training data, and evaluation caveats next to scores?
What metadata is needed to distinguish official code, reproduction attempts, forks, and unrelated repositories?
How can open research-code indexes remain maintained when papers, libraries, and datasets change quickly?