Compare
paperreview.ai and reviewpaper.ai, compared
paperreview.ai is a free research tool from Stanford, and a good one. If you want reviews of the whole paper in your target venue's actual form, from several reviewers plus an area chair, with every point linked to your PDF, that's what we built.
Last checked against official sources on .
What paperreview.ai is
paperreview.ai is a free AI reviewer from the Stanford ML Group. You upload a PDF and an email address. It converts the paper, searches arXiv for related work, writes a review from a template and scores seven dimensions. When the target venue is ICLR it also shows a 1–10 score. Its tech overview reports a 0.42 Spearman correlation with ICLR 2025 reviewers, about the same as two human reviewers agree with each other (0.41). It's a well-built tool and the price is hard to beat.
Its documented limits: only the first 15 pages are analysed, files are capped at 10 MB, papers must be in English, and results arrive by email.
Side by side
From each site's public pages, checked 2026-10-12.
| paperreview.ai | reviewpaper.ai | |
|---|---|---|
| Price | Free | 3 free reviews, then $10/month for 12 credits (~$0.83 per standard review) or $5 for 5 |
| Built by | Stanford ML Group (research project) | Independent product |
| Pages read | First 15 pages, PDF up to 10 MB | The whole PDF, appendices included (up to 30 MB) |
| Venue format | Optional target-venue dropdown (12 venues + Other) | The venue's own review form, field by field, for 29 venues |
| Reviewers | One review from a template | Three reviewer personas plus an area-chair meta-review |
| Score | 1–10 score, shown only when the target venue is ICLR | Each reviewer scores on the venue's scale, plus an expected range and accept / borderline / reject outlook |
| Evidence | Related work searched on arXiv | Every point linked to the passage in your PDF (click to highlight); related-work references via Semantic Scholar / OpenAlex |
| Action list | Within the review text | Ranked "fix before submitting" list and the venue checklist |
| Delivery | Email notification when ready; also viewable on the site | In the browser, usually a few minutes |
| Model | Not user-selectable | Choose Claude Sonnet 5.5, Opus 5.5 or Fable 5.1 |
| Follow-up | — | Chat with the paper, answers cite the page |
| Published evaluation | Yes: Spearman 0.42 with ICLR 2025 reviewers (vs 0.41 human–human) | Not yet |
Which to use when
- Use paperreview.ai if you want a free second opinion, your paper fits in 15 pages, and you're happy to wait for an email.
- Use reviewpaper.ai if the appendix matters (proofs, ablations, extra experiments), you want to see the review in your target venue's form with several reviewers and an area chair, or you want to click each critique and land on the exact passage.
- Use both. Two independent simulated reviews that agree on a weakness are a strong signal.
FAQ
Is my paper used for training?
Not at reviewpaper.ai: papers are stored privately, never used to train models, and you can delete them any time (privacy policy). paperreview.ai's public pages don't describe how uploads are stored or retained, so check with them directly.
Which venues does reviewpaper.ai support?
29 formats, including NeurIPS, ICLR, ICML, CVPR, ACL Rolling Review, AAAI, KDD and TMLR, plus a generic one. Each is explained on its venue page, and venues that don't publish their form are marked as modeled.
How long does a review take?
Usually a few minutes for the Standard model and longer for Deep and Max, which read more carefully.
Try it on your draft
Upload a PDF, pick your venue and get three reviews plus a meta-review in that venue's form. The first three runs are free and need no card.
Review my paper, 3 freeSources
reviewpaper.ai is not affiliated with Stanford, paperreview.ai or any conference or its organizers. Rules change between cycles; the official pages above always win.