Guide
ICML vs NeurIPS: nearly the same review, six months apart
Put the two review forms side by side and they are almost the same document: four rated dimensions, a six-point recommendation, a five-point confidence. What separates these venues in practice is the calendar, not their values.
Published 11 August 2026
Already sent it to one of them? Then the calendar is settled and the reviews are the next thing to read. What the scores mean
The short version
- The two review forms are nearly the same document. Four rated dimensions, a six-point recommendation and a five-point confidence at both venues.
- What separates them is the calendar, not their values. ICML 2026 closed on 28 January, NeurIPS 2026 on 6 May. For most papers one of those is reachable and the other is not.
- The abstract deadline is the one people miss. Both venues close abstract registration days before the paper deadline. Take the current date from the call, not from here.
- A rejection at one leaves real time before the other. Turned down at ICML in April, NeurIPS is a month away, which is long enough to fix something rather than resubmit it in a different font.
The real difference
They are two deadlines, half a year apart
For most papers this settles it. One of these is reachable with the work you have now and the other is not, and no amount of thinking about which venue suits you better changes that.
| ICML 2026 | NeurIPS 2026 | |
|---|---|---|
| Abstract deadline | 23 January 2026 | 4 May 2026 |
| Paper deadline | 28 January 2026 | 6 May 2026 |
| Reviews back | Reviewing closes 12 March, discussion runs from 24 March | 22 July 2026 |
| Author discussion window | 24 March to 7 April 2026 | 27 July to 3 August 2026 |
| Conference | 6 to 11 July 2026, Seoul | December 2026, across Sydney, Atlanta and Paris |
From the ICML 2026 call for papers and the NeurIPS 2026 dates page, read August 2026. Both venues close abstract registration before the paper deadline, and that earlier date is the one people actually miss. Take yours from the current call.
The gap is the point, not the inconvenience
January and May are far enough apart that a paper rejected from one can be genuinely improved before the other, rather than resubmitted unchanged in a different font. If you are turned down at ICML in April, NeurIPS is a month away and that month is real working time.
Waiting for the better fit is usually cheap
The cost of holding a paper for the next deadline is a few months. The cost of submitting it unfinished is the same few months plus a rejection, and reviews written about a version you already knew was not ready.
What a reviewer fills in
The forms are the same shape, down to the scale
This is the part the folklore gets wrong. These are not two different philosophies of review. They are two near-identical rubrics with two dimensions renamed.
| ICML | NeurIPS | |
|---|---|---|
| Rated dimensions, each 1 to 4 | Soundness, presentation, significance, originality. | Quality, clarity, significance, originality. |
| Overall recommendation | 1 to 6. Strong accept, accept, weak accept, weak reject, reject, strong reject. | 1 to 6. Strong accept, accept, borderline accept, borderline reject, reject, strong reject. |
| The only real difference | The middle two are weak accept and weak reject: a judgement about the paper. | The middle two are borderline accept and borderline reject: a judgement about how close the call is. |
| Confidence | 1 to 5, educated guess up to absolutely certain. | 1 to 5, worded the same way. |
From the ICML 2026 reviewer instructions and the NeurIPS 2025 reviewer guidelines. Note that NeurIPS used a 1 to 10 overall score up to and including 2024, so most advice you will find about NeurIPS scores describes a scale that no longer exists.
Soundness and quality, presentation and clarity
Two names for two jobs that are recognisably the same: is the work correct, and can it be read. Both venues score writing quality on its own line, out of 4, apart from whether the science holds. Neither lets a well-argued paper carry a badly written one.
Both split novelty from importance
Originality asks whether it is new. Significance asks whether it matters. A paper can be recorded as novel and unimportant at either venue, which is the failure mode a strong related-work section does nothing about.
Neither has a neutral seat
A six-point scale with no midpoint forces every reviewer onto a side. There is no score that means no opinion, so a reviewer who is unsure still has to lean, and your rebuttal is aimed at the two middle positions where leaning is all that happened.
Two things we will not tell you
Where this page stops
Both are worth stating plainly, because every other page answering this question states them anyway and sources none of it.
Both venues publish their numbers, but ICML makes you hunt for them
NeurIPS puts its figures in a programme chairs' post: 5,290 papers accepted from 21,575 valid submissions in 2025, 24.52%. ICML publishes through its official social accounts and its blog rather than a statistics page, which is why most pages quoting an ICML rate cite nothing. On 30 April 2026 it reported 23,918 submissions entering review and 6,352 accepted, 26.6%. On 7 July 2026 it gave a fuller accounting of the same cycle: 24,661 submissions, 6,552 accepted at the same 26.6%, 4,257 withdrawn and 939 desk rejected. Both sets are ICML's own, they do not quite reconcile, and the honest way to use either is with its date attached.
We have not measured ICML
Our own scoring was benchmarked against 697 real decisions from ICLR and NeurIPS. It has never been checked against an ICML outcome, so nothing we could say about how the ICML bar compares would be evidence. The comparison we can back is between ICLR and NeurIPS, and it is on its own page.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
If the forms are the same, the calendar decides
Both venues rate soundness, writing, novelty and importance separately, out of 4. What differs is the date, so the open question is which rooms are still reachable with the paper you have now.
Before the deadline, the list of things that end a submission without a review is the pre-submission checklist. After the reviews, it is how to write a rebuttal.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.