Read papers
A free way to skim the day’s arXiv: one list per category, with an AI-generated FAQ on each paper to help you triage.
Quick facts
| Pricing | Free |
|---|---|
| Free tier | Yes — browsing, PDF preview and the AI FAQ, with no registration |
| Platforms | Web |
| Developer | Jianlin Su |
| Launched | 2024 |
| Website | https://papers.cool/ |
What Read papers is
Cool Papers is a reading aid for arXiv rather than a reference manager. It mirrors each day’s new submissions, one page per arXiv category, with title, authors, subject tags and the full abstract in a single scrollable list. Each entry carries buttons for the PDF, for copying the citation details to the clipboard, and for a Kimi-generated FAQ. The FAQ is the reason to use it: a fixed set of questions about the problem the paper attacks, the prior work, how it was tested and what remains open, answered from the paper rather than the abstract. That is a relevance filter between skimming and a full read, and it is faster than opening twenty PDFs to find out. The site covers all of arXiv, not only computer science, and shows the latest day with no archive. It stores no library and exports no citations. It was built by Jianlin Su, who writes the Science Space blog.
Best for
- Daily triage of a fast-moving arXiv category
- Deciding whether an abstract deserves a full read
- People who would rather skim ten papers than commit to one
Not ideal for
- Literature management, citation export or annotated libraries
- Searching older work, since only the latest day is online
Pros
- One page per category beats scrolling the raw arXiv listing
- The per-paper FAQ answers are a genuine relevance filter
- Free, no account, nothing to install
Cons
- No history: a missed day cannot be browsed back
- AI answers can be wrong and are no substitute for reading the paper
- No library, tags, notes or citation export
Pricing in detail
Free, with no account and no paid tier. Nothing on the page is gated behind a signup and there is no upsell to a pro version. The AI answers come from an external model rather than one the site hosts, so the practical limit is how that service responds rather than a quota the site publishes. Plan on a cost of zero, and treat any rate limit as unknown rather than generous.