reviewpaper.ai

Reference

Review score scales, compared

A 6 is a strong score at NeurIPS and a borderline one at ICLR. Here is every venue's overall scale, what each number is called, and roughly where the acceptance bar sits.

Last checked against official sources on .

Score converter

Pick a venue and a score. The table shows the score at the same distance from each other venue's rough acceptance bar. It is useful for intuition (“is a NeurIPS 4 like an ICLR 6?”), not for predicting a decision. For ICLR, the acceptance-odds calculator uses real past decisions instead.

ICLR 6 (“marginally above the acceptance threshold”) is above the rough acceptance bar of 5. The same position on other scales:

VenueScaleSame positionNearest label
NeurIPS1–6≈ 4(4.40)Borderline accept
ICML1–6≈ 4(4.40)Weak accept
AISTATS1–7≈ 5(5.40)Borderline Accept
UAI1–10≈ 7(6.80)Accept
TMLR1–4≈ 3(3.20)Leaning Accept
COLT1–5≈ 4(3.80)Accept
ACL ARR1–5≈ 4(3.80)Conference
EMNLP1–5≈ 4(3.80)Conference
NAACL1–5≈ 4(3.80)Conference
COLING1–5≈ 4(3.80)Conference
EACL1–5≈ 4(3.80)Conference
AACL1–5≈ 4(3.80)Conference
COLM1–10≈ 7(6.80)Good paper, accept
CVPR1–5≈ 4(3.80)Weak Accept
ICCV1–6≈ 4(4.40)Borderline Accept
ECCV1–5≈ 4(3.80)Weak Accept
WACV1–5≈ 4(3.80)Accept
AAAI1–8≈ 6(5.60)Weak accept
IJCAI1–9≈ 6(5.80)Weak Accept. Useful. A good paper. The results and insights will benefit the field. I believe it should be accepted.
KDD1–4≈ 3(3.20)Moderate - The paper demonstrates solid technical quality with a sound methodology and thorough analysis. The results are reliable and well-supported. There may be minor issues, but they do not significantly undermine the overall quality. The work is competently executed and meets acceptable standards.
WWW-3–3≈ 1(1.40)weak accept
SIGIR-3–3≈ 1(1.40)weak accept
ICASSP1–4≈ 3(3.00)Marginal accept
Interspeech1–6≈ 4(4.20)Weak Accept
ACM MM1–6≈ 4(4.40)Borderline Accept - I lean towards acceptance, but I could be convinced otherwise.
CoRL1–4≈ 3(3.00)Weak accept
MLSys1–5≈ 3(3.40)Weak accept

A heuristic, not an official equivalence: it keeps a score's distance from each venue's rough acceptance bar. Real decisions depend on the reviews, the confidence scores and the area chair.

All scales at a glance

The overall recommendation each venue asks reviewers for, and the rough bar we use when simulating reviews.

VenueFieldScaleRough barForm
NeurIPSOverall1–6≥ 4 Official form
ICLRRating0–10 (step 2)≥ 5 Modeled
ICMLOverall Recommendation1–6≥ 4 Official form
AISTATSRating1–7≥ 5 Official form
UAIOverall Score1–10≥ 6 Official form
TMLRRecommendation1–4≥ 3 Official form
COLTOverall Score1–5≥ 3.5 Modeled
ACL ARROverall Assessment1–5 (step 0.5)≥ 3.5 Official form
EMNLPOverall Assessment1–5 (step 0.5)≥ 3.5 Official form
NAACLOverall Assessment1–5 (step 0.5)≥ 3.5 Official form
COLINGOverall Assessment1–5 (step 0.5)≥ 3.5 Official form
EACLOverall Assessment1–5 (step 0.5)≥ 3.5 Official form
AACLOverall Assessment1–5 (step 0.5)≥ 3.5 Official form
COLMRating1–10≥ 6 Modeled
CVPRPreliminary Recommendation1–5≥ 3.5 Modeled
ICCVPreliminary Recommendation1–6≥ 4 Modeled
ECCVRating1–5≥ 3.5 Modeled
WACVRating1–5≥ 3.5 Modeled
AAAIOverall Evaluation1–8≥ 5 Modeled
IJCAIOverall Assessment1–9≥ 5 Modeled
KDDTechnical Quality1–4≥ 3 Modeled
WWWOverall Evaluation-3–3≥ 1 Modeled
SIGIROverall Recommendation Score-3–3≥ 1 Modeled
ICASSPOverall Evaluation Of This Paper1–4≥ 2.8 Modeled
InterspeechOverall Recommendation1–6≥ 3.8 Modeled
ACM MMPreliminary Rating1–6≥ 4 Modeled
CoRLRating1–4≥ 2.8 Modeled
MLSysOverall Merit1–5≥ 3 Modeled

Reading your scores

  • The mean is not the decision. Area chairs read the reviews, weigh confidence, and look at whether concerns were answered in the rebuttal. A 3-reviewer mean hides a lot.
  • Look at the spread. One outlier reviewer with low confidence is easier to address than three lukewarm scores.
  • Bars move every year. They depend on the submission pool and the acceptance rate. The bars here are rough, from public statistics where they exist and from each scale's wording where they don't.
  • “Modeled” venues don't publish their form. Their labels are reconstructed from public guidelines, and numbers may differ from what reviewers actually see.
Rebuttal season? See how to write a rebuttal and the ICLR discussion-period guide.

Scale by venue

Every value with its label as reviewers see it, and where the bar sits.

NeurIPS 2026

Official form

Overall

  1. 6Strong Accept
  2. 5Accept
  3. 4Borderline accept
  4. 3Borderline reject
  5. 2Reject
  6. 1Strong Reject

NeurIPS 2025 accepted 24.5% of main-track papers. Accepted papers averaged about 4.3 and rejected about 3.6 (Paper Copilot). Rough rule: mean ≥ 4.0 likely accept, 3.5–4.0 borderline, < 3.5 likely reject.

ICLR 2026

Modeled

Rating

  1. 10strong accept, should be highlighted at the conference
  2. 8accept, good paper
  3. 6marginally above the acceptance threshold
  4. 4marginally below the acceptance threshold
  5. 2reject, not good enough
  6. 0strong reject

ICLR 2026 accepted 27.4% of decided submissions. Paper Copilot means: posters ≈ 5.4, orals ≈ 6.4, rejected ≈ 4.0. Rough rule: mean ≥ 5 likely accept, ~4.4–5 borderline.

ICML 2026

Official form

Overall Recommendation

  1. 6Strong Accept
  2. 5Accept
  3. 4Weak accept
  4. 3Weak reject
  5. 2Reject
  6. 1Strong Reject

ICML 2026 accepted about 26.6% of submissions. No public score statistics; inferred from NeurIPS's identical scale: mean ≥ 4 likely accept, 3.5–4 borderline.

AISTATS 2026

Official form

Rating

  1. 7Strong Accept
  2. 6Accept
  3. 5Borderline Accept
  4. 4Weak Accept / Weak Reject
  5. 3Borderline Reject
  6. 2Reject
  7. 1Strong Reject

The AISTATS 2026 call says acceptance tends to be around 25% (2025: 31%, 2024: 28%). No score statistics are public; on the 1–7 scale a mean of 5 (Borderline Accept) or more is a rough proxy for acceptance, 4–5 borderline.

UAI 2026

Official form

Overall Score

  1. 10Award quality
  2. 9Very Strong Accept
  3. 8Strong Accept
  4. 7Accept
  5. 6Weak Accept
  6. 5Borderline accept. Please use sparingly.
  7. 4Borderline reject. Please use sparingly.
  8. 3Reject
  9. 2Strong Reject
  10. 1Very Strong Reject

UAI historically accepts about 30% of submissions; no 2026 statistics are public. The reviewer instructions target at most 25% of papers at 7 (Accept) or above, so a mean around 6 is competitive and 5–6 is borderline.

TMLR 2026

Official form

Recommendation

  1. 4Accept
  2. 3Leaning Accept
  3. 2Leaning Reject
  4. 1Reject

TMLR has no acceptance quota and publishes no rate. The action editor accepts when both criteria hold (claims supported, some audience interested), so a mean of 3 (Leaning Accept) or more is the rough line. Recommendation scores are mapped 1 Reject … 4 Accept by us; the labels are TMLR's.

COLT 2026

Modeled

Overall Score

  1. 5Strong accept
  2. 4Accept
  3. 3Borderline
  4. 2Reject
  5. 1Strong reject

COLT 2026 accepted about 190 papers; submission counts are not published (historically ~30% acceptance). On this modeled 1–5 scale a mean of ≥ 3.5 is a rough proxy for acceptance.

ACL ARR 2026

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

EMNLP 2026

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

EMNLP has published a Findings of EMNLP volume every year since 2020. In the Overall Assessment, 4 ("Conference") argues for the main conference and 3 ("Findings") for the Findings of the ACL volume, which the program chairs may offer to committed papers that are sound but less exciting. ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

NAACL 2027

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

Recent NAACL editions (2022, 2024, 2025) published a Findings volume. In the Overall Assessment, 4 ("Conference") argues for the main conference and 3 ("Findings") for the Findings of the ACL volume, which the program chairs may offer to committed papers that are sound but less exciting. ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

COLING 2027

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

COLING has not historically published a Findings volume; check the COLING 2027 call. In the Overall Assessment, 4 ("Conference") argues for the main conference and 3 ("Findings") for the Findings of the ACL volume, which the program chairs may offer to committed papers that are sound but less exciting. ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

EACL 2027

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

EACL has published a Findings of EACL volume since 2023. In the Overall Assessment, 4 ("Conference") argues for the main conference and 3 ("Findings") for the Findings of the ACL volume, which the program chairs may offer to committed papers that are sound but less exciting. ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

AACL 2026

Official form

Overall Assessment

  1. 5Consider for Award
  2. 4.5Borderline Award
  3. 4Conference
  4. 3.5Borderline Conference
  5. 3Findings
  6. 2.5Borderline Findings
  7. 2Resubmit next cycle
  8. 1.5Resubmit after next cycle
  9. 1Do not resubmit

AACL-IJCNLP published a Findings volume in 2022 and 2023. In the Overall Assessment, 4 ("Conference") argues for the main conference and 3 ("Findings") for the Findings of the ACL volume, which the program chairs may offer to committed papers that are sound but less exciting. ARR publishes no cutoff; venue senior ACs decide. Rule of thumb: average Overall Assessment ≥ 3.5 → main conference plausible, 3.0–3.5 → Findings, < 3.0 → likely rejected. Main conference acceptance is about 20–25%, plus ~13–15% Findings.

COLM 2026

Modeled

Rating

  1. 10Top 5% of accepted papers, seminal paper
  2. 9Top 15% of accepted papers, strong accept
  3. 8Top 50% of accepted papers, clear accept
  4. 7Good paper, accept
  5. 6Marginally above acceptance threshold
  6. 5Marginally below acceptance threshold
  7. 4Ok but not good enough - rejection
  8. 3Clear rejection
  9. 2Strong rejection
  10. 1Trivial or wrong

COLM accepted 28.9% in 2024. Accepted papers averaged about 6.4–6.5 on the 1–10 rating in 2024 and 2025 (Paper Copilot). Rough rule: mean ≥ 6 competitive, 5.25–6 borderline.

CVPR 2026

Modeled

Preliminary Recommendation

  1. 5Strong Accept
  2. 4Weak Accept
  3. 3Borderline
  4. 2Weak Reject
  5. 1Reject

CVPR 2026 accepted about 25% of 16,092 submissions. No score statistics are public; on this 1–5 mapping, a mean of ≥ 3.5 (mostly Weak Accept or better) is a rough proxy for acceptance.

ICCV 2025

Modeled

Preliminary Recommendation

  1. 6Accept
  2. 5Weak Accept
  3. 4Borderline Accept
  4. 3Borderline Reject
  5. 2Weak Reject
  6. 1Reject

ICCV 2025 accepted 2,701 of 11,239 valid submissions (24%, "not pre-determined"). No score statistics are public; on this modeled 1–6 scale a mean of ≥ 4 (Borderline Accept) is a rough proxy for acceptance.

ECCV 2026

Modeled

Rating

  1. 5Strong Accept
  2. 4Weak Accept
  3. 3Borderline
  4. 2Weak Reject
  5. 1Reject

ECCV 2026 accepted about 27.5% of ~10,500 submissions (secondary sources; ECCV 2024: 27.9%). No score statistics are public; on this modeled 1–5 scale a mean of ≥ 3.5 is a rough proxy for acceptance.

WACV 2027

Modeled

Rating

  1. 5Strong Accept
  2. 4Accept
  3. 3Neutral
  4. 2Reject
  5. 1Strong Reject

WACV 2026 accepted 832 of 2,550 submissions (33%): round 1 38% (84 direct accepts, 439 of 736 revisions accepted), round 2 26%. No score statistics are public; on this modeled 1–5 scale a mean of ≥ 3.5 is a rough proxy for acceptance or a revise-and-resubmit invitation.

AAAI 2026

Modeled

Overall Evaluation

  1. 8Strong accept
  2. 7Accept
  3. 6Weak accept
  4. 5Borderline accept
  5. 4Borderline reject
  6. 3Weak reject
  7. 2Reject
  8. 1Strong reject

AAAI-26 accepted 17.6% of 23,680 submissions, and Phase 1 removes many papers before author feedback. No public score statistics; rough rule: mean ≥ 5 (of 8) needed to be competitive.

IJCAI 2026

Modeled

Overall Assessment

  1. 9Award Quality
  2. 8Strong Accept. Inspiring. An excellent paper. This paper changed my thinking on this topic. I will fight for acceptance.
  3. 7Clear Accept. Interesting. A very good submission. I learned a lot from this paper. I vote and argue for acceptance.
  4. 6Weak Accept. Useful. A good paper. The results and insights will benefit the field. I believe it should be accepted.
  5. 5Borderline Accept. Marginally above the acceptance threshold. Technically correct, but not particularly exciting or inspiring. Could be accepted more or less in its current form. Not a big loss if it is not included in the program. Please use sparingly.
  6. 4Borderline Reject. Marginally below the acceptance threshold. The paper has merits but there are key weaknesses. It would benefit from another revision. Can be rejected. But having it in the program would not be that bad. Please use sparingly.
  7. 3Weak Reject. Not good enough. Limited contribution. Needs more work and/or revisions. I believe it should be rejected.
  8. 2Clear Reject. Unclear contribution. Needs to be reworked and/or major revision. I vote and will argue for rejection.
  9. 1Strong Reject. Contribution is flawed. I will strongly argue for rejection.

IJCAI-ECAI 2026 accepted about 13.5% of main-track papers (secondary source; 2025: 19.3%). A summary-reject phase removes papers whose two Phase-1 reviews are not positive enough, and a rebuttal rarely rescues reject-leaning reviews. On the 1–9 scale a mean of ≥ 5 (Borderline Accept) is needed to be competitive; most accepts average 6+.

KDD 2026

Modeled

Technical Quality

  1. 4High - The paper exhibits a high level of technical quality with a rigorous and well-executed methodology and analysis. The results are highly reliable, well-supported, and thorough. The work demonstrates technical excellence and sets a high standard for quality in the field.
  2. 3Moderate - The paper demonstrates solid technical quality with a sound methodology and thorough analysis. The results are reliable and well-supported. There may be minor issues, but they do not significantly undermine the overall quality. The work is competently executed and meets acceptable standards.
  3. 2Low - The paper has several technical weaknesses, such as minor methodological flaws, insufficient analysis, or unsupported conclusions. While the work shows some level of competence, it lacks thoroughness and precision. Improvements are necessary for it to be considered robust.
  4. 1Very Low - The paper has major technical flaws that invalidate its main conclusions.

KDD 2026 Cycle 1 accepted 256 of 1,215 research-track papers (21%; KDD 2025: 18.7%). KDD has no overall score visible to authors; Novelty and Technical Quality act as the headline ratings. Shared averages of accepted papers cluster around 3 (Moderate) on each 1–4 axis, so a Technical Quality mean ≥ 3 with Novelty ≥ 3 is competitive and ~2.6–3 borderline.

WWW 2027

Modeled

Overall Evaluation

  1. 3strong accept
  2. 2accept
  3. 1weak accept
  4. 0borderline paper
  5. -1weak reject
  6. -2reject
  7. -3strong reject

WWW 2026 accepted about 676 of 3,370 research submissions (~20%). No score statistics are public; on the −3…3 scale a mean of ≥ 1 (weak accept) is a rough proxy for acceptance, 0.25–1 borderline.

SIGIR 2026

Modeled

Overall Recommendation Score

  1. 3strong accept
  2. 2accept
  3. 1weak accept
  4. 0borderline
  5. -1weak reject
  6. -2reject
  7. -3strong reject

SIGIR full papers are historically accepted at about 20–21%; 2026 figures are not published. On the −3…3 scale a mean of ≥ 1 (weak accept) is a rough proxy for acceptance, 0.25–1 borderline.

ICASSP 2026

Modeled

Overall Evaluation Of This Paper

  1. 4Definite accept
  2. 3Marginal accept
  3. 2Marginal reject
  4. 1Definite reject

ICASSP 2026 accepted close to 4,800 of ~11,200 regular submissions (~43%). TPC chairs set the rate; no thresholds are published. On the 1–4 overall scale a mean of ≥ 2.75 (mostly Marginal accept or better) is a rough proxy for acceptance.

Interspeech 2026

Modeled

Overall Recommendation

  1. 6Definite Accept
  2. 5Accept
  3. 4Weak Accept
  4. 3Weak Reject
  5. 2Reject
  6. 1Definite Reject

Interspeech accepts roughly 50% of regular papers (official special-sessions call); the new long-paper track targets below 30%. On this 1–6 scale a mean of ≥ 3.75 (Weak Accept or better) is a rough proxy for acceptance.

ACM MM 2026

Modeled

Preliminary Rating

  1. 6Accept - This submission should definitely be presented at ACM Multimedia.
  2. 5Weak Accept - I would need strong arguments to reject this submission.
  3. 4Borderline Accept - I lean towards acceptance, but I could be convinced otherwise.
  4. 3Borderline Reject - I lean towards rejection, but I could be convinced otherwise.
  5. 2Weak Reject
  6. 1Reject

ACM MM 2025 accepted 1,250 of 5,330 research papers (23.5%; 2024: 26.2%). No score thresholds are public; on the 1–6 scale a mean of ≥ 4 (Borderline Accept) is a rough proxy for acceptance.

CoRL 2026

Modeled

Rating

  1. 4Strong accept
  2. 3Weak accept
  3. 2Weak reject
  4. 1Strong reject

CoRL 2026 accepted roughly a third of ~2,100 submissions (secondary source; 2024: ~39%). Papers need at least one weak accept to survive the first round. On this modeled 1–4 scale a mean of ≥ 2.75 is a rough proxy for acceptance.

MLSys 2026

Modeled

Overall Merit

  1. 5Strong accept
  2. 4Accept
  3. 3Weak accept
  4. 2Weak reject
  5. 1Reject

MLSys 2026 published about 135 papers; acceptance rates are not published (historically ~20–25%). On this modeled 1–5 scale a mean of ≥ 3 (Weak accept) is a rough proxy for acceptance.

See the score your draft would get

Three simulated reviewers score your paper on your target venue's own scale, an area chair writes the meta-review, and you get the expected score range plus a ranked list of fixes. Your first three runs are free.

Review my paper, 3 free

Sources

reviewpaper.ai is not affiliated with any of these venues or its organizers. Rules change between cycles; the official pages above always win.