NeurIPS - the Conference on Neural Information Processing Systems - is the largest annual conference in machine learning and artificial intelligence, and one of the most competitive publication venues in all of computer science. First held in Denver, Colorado in 1987 and renamed from NIPS to NeurIPS in November 2018, the conference is run by the Neural Information Processing Systems Foundation, a US non-profit whose board consists largely of previous general chairs. In computer science, top conference proceedings function as primary publications: a NeurIPS paper is a full peer-reviewed, archival research contribution, not an abstract or a poster summary.

For 2025, the main track received 21,575 submissions and accepted 5,290 papers (about 24.5%), with the conference running in both San Diego and Mexico City. This guide covers how the venue works: which track fits your paper, the annual submission cycle, OpenReview peer review, formatting requirements, and what publication costs (nothing).

NeurIPS runs on a fixed annual calendar rather than rolling submission. The 2025 cycle is representative:

Miss the deadline and the next opportunity is a full year away - many authors treat NeurIPS, ICML, and ICLR as one staggered submission calendar for the same community.

All reviewing runs on OpenReview. Key features of the model:

• Double-anonymized at the reviewer and area-chair level: submissions must contain no identifying information, and reviewer identities are never revealed. Senior area chairs and program chairs can see author identities.
• Rebuttal and discussion: authors get a one-week window to respond to reviews, followed by structured reviewer-author and reviewer-AC discussion periods.
• Public reviews after decisions: reviews, meta-reviews, and author discussions for accepted papers are made public on OpenReview (with reviewer identities kept anonymous). Authors of rejected papers can opt in to making theirs public too.
• During review, submissions are visible only to the assigned program committee - NeurIPS does not post submissions publicly before decisions.

There is no fee to submit to NeurIPS and no fee to publish an accepted paper. The complete proceedings - every volume of Advances in Neural Information Processing Systems back to 1987 - are free to read at proceedings.neurips.cc. Authors retain rights to their work, and posting preprints to arXiv before and during review is standard practice in the community.

The one real cost: at least one author of an accepted paper must register for the conference to present it, and registration plus travel is typically the largest expense. The foundation offers financial assistance programs for attendees who need support.

Formatting is strict. Submissions must use the official NeurIPS LaTeX style file for the current year - papers violating the style requirements may be rejected without review. The main text is limited to nine content pages including figures and tables, with references and the mandatory paper checklist not counting against the limit; everything goes in a single PDF (max 50 MB). SciSpace hosts the NeurIPS template so you can write and format your manuscript without wrestling with the class file.

Cite NeurIPS papers as conference proceedings: author(s), title, "Advances in Neural Information Processing Systems" (volume number), year, and page or paper number. The proceedings site provides BibTeX entries for every paper. In-text references use the numbered style defined by the NeurIPS style file.

NeurIPS maintains a code of conduct and a code of ethics for submissions; papers can be rejected on ethical grounds and flagged submissions receive dedicated ethics review. The paper checklist requires authors to address reproducibility, licensing of assets, societal impact, and human-subject considerations. LLM usage in papers must comply with the current-year LLM policy stated in the call for papers.