Guide
ICLR vs NeurIPS: which bar your paper actually clears
The papers NeurIPS turns down score where ICLR's accepted papers sit. Neither conference publishes how its bar compares with the other's, so we put 494 of their real decisions on one scale and looked.
Published 8 August 2026. Measured on ICLR 2025 and NeurIPS 2023, 494 papers, split near half accepted and half rejected by design. Nothing here is an acceptance rate.
Still choosing? Paste your abstract lower on this page and see which venues publish work like it. Jump to the matcher
The short version
- NeurIPS sits higher on both sides of its own line. Median accepted paper 6.90 against ICLR's 6.70, median rejected 6.03 against 5.75, across 494 real decisions read by the same weighting.
- The overlap is where most papers actually live. The strongest quarter of what NeurIPS rejected sits at 6.40; the weakest quarter of what ICLR accepted sits at 6.45. Five hundredths apart.
- A NeurIPS rejection tells you less than an ICLR one. 11 of 97 NeurIPS rejects score at or above the median accepted ICLR paper, against 7 of 149 the other way round.
- None of this is an acceptance rate. It is where papers sat when one reader scored all of them the same way, which is the comparison neither conference publishes.
The bar
NeurIPS sits higher on both sides of its own line
Same specialists, same weighting, both venues. The scores below never learn where a paper was sent, so the two columns can be read against each other.
| Group | Papers | Lower quartile | Median | Upper quartile |
|---|---|---|---|---|
| NeurIPS, accepted | 100 | 6.55 | 6.90 | 7.30 |
| ICLR, accepted | 148 | 6.45 | 6.70 | 6.93 |
| NeurIPS, rejected | 97 | 5.56 | 6.03 | 6.40 |
| ICLR, rejected | 149 | 5.10 | 5.75 | 6.17 |
The scores are ours, recomputed from result data and produced by the same weighting at both venues. Accepted and rejected are each conference's own decision. NeurIPS runs higher than ICLR at every decile, in both groups.
The two distributions overlap where it counts
The strongest quarter of papers NeurIPS turned down sits at 6.40. The weakest quarter of papers ICLR accepted sits at 6.45. Five hundredths apart, which puts a large band of papers on either side of the line depending only on where they were sent.
One rejection tells you less than the other
11 of the 97 papers NeurIPS rejected score at or above the median accepted ICLR paper. On the ICLR side the same count is 7 of 149. A NeurIPS rejection is more than twice as likely to have landed on a paper scoring in ICLR's accepted range.
NeurIPS sorts just as cleanly, it starts higher
Both venues split their accepts from their rejects by about nine tenths of a point. The sorting power is the same at each. The line is drawn higher at one.
What each one asks for
Both hand in numbers, and they are not the same numbers
It is widely repeated that ICLR reviewers write prose while NeurIPS reviewers fill in scores. That is not true: both file dimensional ratings. What differs is which dimensions, how the overall score is shaped, and who gets to read the result.
| ICLR | NeurIPS | |
|---|---|---|
| The rated dimensions | Soundness, presentation and contribution, each 1 to 4, plus confidence. | Quality, clarity, significance and originality, each 1 to 4, plus confidence. |
| How novelty is scored | Folded into contribution along with importance, so one number carries both. | Split in two. Originality rates whether it is new, significance rates whether it matters, and a paper can be recorded as novel and unimportant. |
| The overall score | A gapped set. Through 2025 it was 1, 3, 5, 6, 8, 10, with 5 and 6 named as just below and just above the acceptance threshold. For 2026 it is mapped to 0, 2, 4, 6, 8, 10. | Contiguous 1 to 6 since 2025, strong reject to strong accept. Earlier editions used 1 to 10, and most advice online still describes that one. |
| Who reads the record | Everyone. Reviews, your replies and the reviewers' replies are public on OpenReview, and stay public whatever the decision. | Only the people assigned. A thin review costs its author nothing in public, and your reply is seen by the chair and the reviewers. |
| Rebuttal | A public discussion phase. Across ICLR 2024 and 2025, most review scores did not move at all, and the papers whose scores did rise were accepted at 56 to 58 percent against 8 to 12 percent for those that stayed flat. | A rebuttal exists, and the dimensional scores are what it has to move. |
Fields and scales from each venue's published guidelines and review forms, read August 2026: NeurIPS 2025 and ICLR, whose scale lives in the OpenReview form rather than the guide. The ICLR fields, both anchor sets and the rebuttal figures come from Kargaran et al., a study of over 19,000 ICLR papers. Both venues retune these between editions, so check the current form.
Choosing
Which one to send it to
NeurIPS, if the work is new but the importance is arguable
Originality and significance are rated separately there, so being first is recorded on its own line rather than being averaged away by a reviewer who doubts the work matters yet. At ICLR both collapse into one contribution score, and a reviewer unconvinced by the importance marks the novelty down with it.
ICLR, if the idea is ahead of the evidence
The line sits lower and the discussion phase is real and public. A contribution that needs arguing has somewhere to be argued, in front of the area chair, with the reviewer's original wording still on the page next to your reply.
Turned down at NeurIPS? ICLR is a real second room
About one NeurIPS rejection in nine lands on a paper scoring at or above the median ICLR accept, twice the rate on the ICLR side. Before you rewrite the paper, work out whether it was only in the wrong room.
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.
The table stops at two venues
The matcher ranks every venue we hold against your abstract, not only these two, and it reads the paper rather than the reputation. Free, no account, and nothing you paste is kept.
What this kind of scoring cannot tell you is set out in what AI peer review cannot do. The objections that end papers at either venue, sorted by whether they can still be fixed, are in why papers get rejected.
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.