Decision in 20 seconds
If you need the best places to track AI research papers, keep the routine simple: use arXiv or Hugging Face Papers to notice new work, use Papers with Code or OpenReview to judge whether it matters, and only read deeply when the paper changes what you may benchmark, test, or ship.
Key points
- Paper discovery is only useful if it helps you decide what to read, test, or ignore.
- The strongest stack separates first notice, benchmark context, and deeper review rather than expecting one source to do everything.
- For builder workflows, the question is not 'what paper exists?' but 'which paper changes what we should benchmark, prototype, or watch next?'
What changed recently
- Last reviewed: 2026-05-10.
- This shortlist now makes the discovery layer explicit and adds clearer source roles for paper tracking, evaluation, and follow-up reading.
Explanation
Most teams do not need to read every new AI paper. They need a stable routine that catches the papers most likely to affect evaluation, product direction, or tool choices.
That is why this page prioritizes sources that surface papers early and make it easier to connect them to code, benchmarks, and community review.
How we picked this shortlist
A research-paper source makes the shortlist when it helps you discover work early, verify what the paper actually claims, and decide whether it deserves a deeper read or practical test.
| Source type | Best for | Why it matters | Limit |
|---|---|---|---|
| arXiv | First notice | It is the fastest broad discovery layer for newly posted research papers | Raw volume is high and not every paper is practically relevant |
| Papers with Code | Benchmark context | It helps connect papers to tasks, code, and leaderboard movement | Not every paper or task is covered equally well |
| OpenReview | Conference review context | Useful when you want to see review-stage discussion or accepted conference work | Coverage depends on venue and timing |
| Hugging Face Papers | Curated reading flow | Useful for a lighter paper-reading routine tied to a familiar AI workflow stack | It is still a discovery layer, not the primary research archive |
How to verify the answer
Research-paper tracking works best when each source has one job: archive, benchmark context, review context, or curated discovery. Use that separation to keep the workflow clear.
Tools / Examples
- Use the evidence timeline to verify claims quickly.
- Follow the sources section for primary-source citation.
Evidence timeline
The primary discovery layer for newly posted papers and the fastest way to see what entered the public research stream.
Useful when you need benchmark context, task mapping, and code links rather than paper titles alone.
Useful for following conference-linked work and understanding review-stage or venue-specific context.
Useful as a lighter, more curated paper-discovery layer for teams already centered on model and tooling workflows.
Sources
- arXiv
- Papers with Code
- OpenReview
- Hugging Face Papers
- RadarAI Methodology
- Sources & Coverage
- Signals Library
FAQ
How is this page maintained?
It is updated when new evidence appears, rather than creating thin pages for every headline.
How should I cite this page?
Use the primary source links for any citation or decision; cite this page as a summary layer if needed.
Search angles this page supports
track AI research papers research papers arXiv Papers with Code
Related
Go deeper
- Best way to track AI evals and benchmarks
- AI monitoring workflow for builders
- How to verify AI news sources
Last updated: 2026-06-23 · Policy: Editorial standards · Methodology