ICLR 2027 · score distribution

How AI reviewers score every ICLR 2027 submission

We reviewed every public ICLR 2027 submission ten independent times with GPT-6-Luna, using the same fixed prompt, and averaged the ten scores. This page shows the distribution across all submissions, how the tracks differ, and where any score would rank.

To avoid influencing the ongoing review process, per-paper results will be released after the ICLR 2027 review period ends. For now we publish the distribution only. You can review your own paper ten times with GPT-6-Luna, ask for a non-integer score each time, and enter the average below to see where it would rank.

Score your own paper with our prompt. Copy it into Codex on your computer and attach your PDF; it prepares the paper and reviews it the same way we did. Scores vary a lot from run to run, so treat the result as a rough reference. The repositories and tools mentioned in the prompt are public; HackMyPaper does not hold their copyright.

Where would your score rank?

Free

Ranks are estimated over all submissions; ties share a rank. The closer your score follows our setup (GPT-6-Luna, ten reviews, averaged), the more meaningful the comparison.

Distribution of scores across all submissions

Mean of ten reviews per paper, 0.5-point bins

Tracks at a glance

Sorted by median

10th–90th percentileMiddle 50%MedianOverall 25% cut-off

Track comparison

Click a column to sort. “Top-25% cut-off” is the score a paper needs to be in the top quarter of its own track.

By submission status

Distribution within each track

Coverage. Every public submission is included except a handful whose PDFs could not be processed.

AI scores cannot replace human evaluation. On our ICLR 2025/2026 sample, the strongest AI model we tested predicts acceptance with an AUC of about 0.72; for comparison, page count alone reaches 0.62. This page only shows how AI reviewing sees the field.

Want a review of your own paper? The reviews here use GPT-6-Luna so that every submission could be reviewed ten times. Reviews you order on HackMyPaper run on GPT-6-Sol with medium reasoning and can be repeated up to 10 times (Full review). Every new account gets 5 free credits, enough for one Standard review.

Try it on your paper →

Reviews are simulated by an AI model and are not ICLR reviews. They carry no information from the official review process and are provided for research on automated review.