ICLR 2027 acceptance odds: what do your review scores mean?
Enter your reviewers' ratings and see what happened to papers with similar scores at ICLR 2024, 2025 and 2026, from 37,165 reviewed submissions on OpenReview. Free, no sign-up.
Your reviews
ICLR 2026 / 2027 scale- R1Reviewer 1
- R2Reviewer 2
- R3Reviewer 3
- R4Reviewer 4
0 strong reject · 4 marginally below · 6 marginally above · 10 strong accept
Try a profile
Compare against
Same 0–10 even-number scale as ICLR 2027 (assumed until ICLR confirms it).
Estimated acceptance odds
Papers with scores like 6 / 6 / 8 / 4 at ICLR 2026 that stayed in to a decision
94%
accepted · likely range 92%–95%
Historically, almost always accepted
- Mean rating
- 6.00
- Percentile of mean
- 93th
- Oral
- 5.1%
- Similar papers withdrawn
- <1%
- Effective sample
- ~641 papers
All 1,012 ICLR 2026 papers with a mean of 5.75–6.25:
- Oral49 (5%)
- Poster887 (88%)
- Rejected68 (7%)
- Withdrawn8 (1%)
Same scores against each year
Acceptance rate by mean initial rating, ICLR 2026
Share of decided papers that were accepted, in bins of 0.5. The band is a 90% interval; hollow dots have fewer than 12 decided papers. Grey bars count every reviewed paper in the bin, withdrawn ones included. Hover a bin for numbers.
What happened to papers with these exact scores
The most common 4-reviewer profiles at ICLR 2026 (initial ratings, order doesn't matter), and yours if it isn't among them.
| Initial ratings | Mean | Papers | Accepted* | Oral | Poster | Rejected | Withdrawn |
|---|---|---|---|---|---|---|---|
| 4 / 2 / 2 / 2 | 2.50 | 711 | <1% | 0 | 1 | 273 | 437 |
| 4 / 4 / 2 / 2 | 3.00 | 1,080 | <1% | 0 | 4 | 446 | 630 |
| 6 / 4 / 2 / 2 | 3.50 | 907 | 3% | 0 | 15 | 460 | 432 |
| 4 / 4 / 4 / 2 | 3.50 | 960 | 4% | 0 | 18 | 446 | 496 |
| 6 / 6 / 2 / 2 | 4.00 | 296 | 15% | 0 | 31 | 174 | 91 |
| 6 / 4 / 4 / 2 | 4.00 | 1,503 | 11% | 0 | 113 | 922 | 468 |
| 4 / 4 / 4 / 4 | 4.00 | 416 | 12% | 0 | 29 | 219 | 168 |
| 6 / 6 / 4 / 2 | 4.50 | 923 | 32% | 1 | 256 | 550 | 116 |
| 6 / 4 / 4 / 4 | 4.50 | 1,061 | 24% | 0 | 209 | 672 | 180 |
| 8 / 6 / 4 / 2 | 5.00 | 416 | 50% | 1 | 194 | 195 | 26 |
| 6 / 6 / 4 / 4 | 5.00 | 1,093 | 55% | 4 | 573 | 471 | 45 |
| 8 / 6 / 4 / 4 | 5.50 | 490 | 78% | 7 | 368 | 103 | 12 |
| 6 / 6 / 6 / 4 | 5.50 | 598 | 84% | 7 | 488 | 93 | 10 |
| 8 / 6 / 6 / 4 (you) | 6.00 | 475 | 95% | 26 | 421 | 25 | 3 |
*Accepted as a share of papers that got a decision (withdrawn papers excluded). Small rows are noisy: the headline estimate pools similar profiles instead.
Read this before you trust the number
- Historical, not predictive. It says how often papers with similar initial ratings were accepted in past years. It has not read your paper or your reviews.
- The area chair decides. ICLR decisions follow the meta-review and the discussion, not the mean. A sharp rebuttal can rescue a 4/4/6; one unanswered fatal flaw can sink an 8/6/6.
- Conditional on staying in. Rates exclude withdrawn papers. Many low-scoring papers withdraw, which is why ICLR's official acceptance rate is lower than the rates shown here.
- ICLR 2027 may differ. The 2027 rating scale, reviewer pool, acceptance rate and process (2026 reset scores after a security incident) can all shift the curve. Compare across years to see how much it moves.
How the estimate works
Every reviewed ICLR paper from that year is weighted by how close its initial ratings are to yours on three things: the mean, the lowest rating and the spread. So 6/6/6 and 8/8/2 have the same mean but different neighbours. The estimate is the weighted share of those papers that were accepted, among the ones that reached a decision.
How quickly the weight falls off was picked by cross-validation for each year. A weak prior from the mean alone keeps rare profiles (say 10/10/0) from jumping to 0% or 100%. The likely range is a 90% interval for that rate given the effective number of similar papers; the year-to-year spread is a better sense of real uncertainty.
Held-out quality (5-fold CV, decided papers): ICLR 2026 AUC 0.89, Brier 0.128 · ICLR 2025 AUC 0.87, Brier 0.143 · ICLR 2024 AUC 0.88, Brier 0.138.
The 2026 scale change
| Label | 2024–25 | 2026–27 |
|---|---|---|
| Strong reject | 1 | 0 |
| Reject, not good enough | 3 | 2 |
| Marginally below the threshold | 5 | 4 |
| Marginally above the threshold | 6 | 6 |
| Accept, good paper | 8 | 8 |
| Strong accept | 10 | 10 |
The bottom half of the scale moved down by one, so a 2026 mean of 5.0 is not a 2025 mean of 5.0. When you compare against 2024 or 2025 we translate each rating by its label.
Borderline? The rebuttal is where it's decided.
Get three practice ICLR reviews and an area-chair meta-review of your paper in the ICLR form, with every point linked to the passage it is about. Then chat with your paper to draft rebuttal answers to your real reviewers' points, grounded in what the paper actually says.
Questions
Is a 6/6/6 accepted at ICLR?
Usually, but not always. At ICLR 2026, 96% of papers with three "marginally above" ratings (or a very similar profile) were accepted once they reached a decision; 75 of the 79 decided papers with exactly 6/6/6 were accepted. At ICLR 2025, where 6 also meant "marginally above", the rate was 95%. Unanimous borderline-positive reviews leave the decision to the area chair, so the meta-review and the discussion matter more than the mean.
What average score gets accepted at ICLR?
There is no fixed cutoff. On initial (pre-rebuttal) ratings, the acceptance rate crossed 50% at a mean of about 5.0 at ICLR 2026 (0–10 even scale) and about 5.3 at ICLR 2025 (1/3/5/6/8/10 scale); it passed 80% around 5.6 and 6.0 respectively. Overall, 39% of decided ICLR 2026 papers were accepted (27.4% of all valid submissions, because many papers were withdrawn or desk-rejected).
What changed in the ICLR rating scale in 2026?
ICLR 2022–2025 used ratings 1, 3, 5, 6, 8, 10. ICLR 2026 switched to even numbers 0, 2, 4, 6, 8, 10 with the same labels: strong reject, reject, marginally below, marginally above, accept, strong accept. The 'marginally below' rating moved from 5 to 4 and 'reject' from 3 to 2, so means are lower on the new scale and are not directly comparable. This calculator compares each year on its own scale and translates your ratings by label. ICLR 2027 has not published its form yet; we assume it keeps the 2026 scale.
Do ICLR scores change after the rebuttal?
Often. At ICLR 2025, 55% of papers had at least one reviewer change their rating during the discussion, 25% of individual ratings changed, and the average paper's mean moved by +0.37. Rebuttals mostly move scores up. ICLR 2026 is the exception: after an OpenReview security incident, all scores were reset to their pre-rebuttal values and new area chairs weighed the discussion instead.
How accurate is this calculator?
It describes what happened to similar papers; it does not know your paper. In 5-fold cross-validation on ICLR 2026, the estimate ranked accepted papers above rejected ones 89% of the time (AUC 0.89), so scores explain a lot but far from everything: area chairs, reviewer discussion, confidence and the content of the reviews decide the rest.
Where does the data come from?
From 37,165 reviewed ICLR 2024–2026 submissions on OpenReview (ratings and decisions are public under CC BY 4.0), with pre-rebuttal ratings for 2024 and 2025 from Kargaran et al.'s ICLR rebuttal dataset (CC BY 4.0). We only keep aggregated score profiles, no titles or reviewer data.
More on the review form itself: the ICLR review form, field by field · ICLR 2027 dates