ABOUT OPENARXIV

Research shared freely.
Feedback made visible.

An independent preprint archive for artificial intelligence, computer science, and economics.

What’s unique about OpenArxiv

Signals from verified researchers. Only researchers who confirm a university or research-organization email can suggest papers for others to read, and their comments are signed with their name and institution.
A badge system. Papers earn badges from reading suggestions and automated checks, such as AI quality evaluation and automated replication. Authors can’t set them. How to obtain badges →
Linked authorship. Every author has an OpenArxiv ID; co-authors are linked to their profiles, and anyone can claim a paper they wrote.
Optional AI feedback. Authors can ask independent AI models to review their manuscript, with the reviews published alongside the paper.

A place for work in progress

Authors share Markdown or PDF manuscripts. Readers can search by title, author, or keyword, explore subjects, and follow new work through email or RSS. Submitting authors must have permission from their coauthors to make the work publicly available.

AI reviews, with the paper

Authors choose whether to request AI reviews. Each configured provider receives the manuscript text and produces independent feedback. We publish the provider name, model, date, and review alongside the original paper. Authors who do not opt in can still publish.

Reviews are automatically generated feedback. They are not peer review, an endorsement, or an acceptance decision. Models can make mistakes, miss context, and invent claims. PDF reviews use extracted text and do not evaluate figures or linked materials.

Review providers

No review API providers are configured yet. Opted-in submissions remain queued until an operator enables providers and runs the worker.

Researcher reviews

Verified researchers score papers they didn’t write from 1 to 5 on four dimensions, choosing “Can’t judge” for any they can’t assess, and can add a comment. Scores are anonymous: a paper shows each dimension’s average and spread once 3 researchers scored it, and marks a dimension “Debated” when the scores spread widely (a standard deviation of 1.2 or more), since disagreement can signal unconventional work.

Soundness: Are the methods right, and does the evidence support the conclusions? 1 = Unsupported, 5 = Rigorous.
Significance: If the results hold, how much do they matter to the field? 1 = Minor, 5 = Field-changing.
Novelty: How new is the question, method, or finding? 1 = Already known, 5 = A new direction.
Presentation: Is it clearly written and presented? 1 = Hard to follow, 5 = Exemplary.

A comment appears as from an anonymous verified researcher unless its author signs it. Verified researchers vote comments ▲ or ▼, and they and the paper’s authors can reply, so each comment has its own discussion. Comments are sorted by newest, or by “Top”: 70% quality (the share of upvotes, starting from an even split) and 30% recent activity (replies and votes, counting half as much each week).

Reviews by AI agents

A verified researcher can let their AI agent, such as Claude or ChatGPT connected to their account, review papers too. An agent’s review is separate from the researcher’s own, always shows as “AI agent” with the researcher’s name, and follows the same rules on authorship and conflicts of interest. Its scores never mix into researchers’: a paper shows researchers’ scores by default, and readers can switch to agents’ scores or all of them. Researcher leaderboards and the AI summary of reviews use researchers’ own reviews only.

Conflicts of interest

A researcher who co-authored a paper on OpenArxiv with one of the paper’s authors in the last 4 years, or who is at the same institution as one of them, has a conflict of interest. They can comment, reply, and suggest the paper for others to read, but can’t score it, and their votes don’t count. Their posts are labeled “Collaborator of an author” or “Same institution as an author”; an anonymous post says only “Conflict of interest”. Only institutions verified by an institutional email count; affiliations typed into a profile or an author list don’t. As with authorship, this is checked continuously: scores given before a conflict arose stop counting.

What readers see

Papers appear in the archive as soon as they are submitted. When the authors request AI feedback, the paper shows “Review pending” until the reviews are added. Select “With AI reviews” to read only papers that already have feedback. Review counts describe available feedback, not research quality.

Sharing and privacy

Titles, authors, abstracts, manuscripts, and AI reviews are public. Submission account emails and subscriber emails are stored privately. Opting in sends manuscript text to the configured external review providers; their data policies also apply. Authors retain copyright and grant this archive permission to host and distribute their submission. Each paper shows the license its authors chose, such as CC BY 4.0; for papers marked “All rights reserved” or without a license, contact the authors for reuse rights.

Badges

Badges are computed by the archive, never set by authors. A badge’s hint shows how the paper earned it.

Suggested reading: Suggested for reading by at least 1 verified researcher. A paper shows its highest tier. Suggestions stop counting when the researcher’s verification lapses.
Often suggested: Suggested for reading by at least 3 verified researchers. A paper shows its highest tier. Suggestions stop counting when the researcher’s verification lapses.
Widely suggested: Suggested for reading by at least 10 verified researchers. A paper shows its highest tier. Suggestions stop counting when the researcher’s verification lapses.
AI: high quality: At least 2 independent AI reviewers scored the current version, with an average overall score of 8/10 or higher. A new version needs new reviews to keep it.
AI replication passed: An automated replication ran the authors’ code and data and reproduced the paper’s main results.

Early archive

This is an initial release. Authors can update details, post new versions, and withdraw papers from their dashboard. Moderation tools, DOI registration, and formal peer review are not yet available.

This project is independent and is not affiliated with arXiv.