The average score threshold for ICML 2026 acceptance will be higher than last year, likely settling around 6.3 to 6.5. That is our read of the room, based on the steady climb in submission quality and the pressure on reviewers to differentiate signal from noise.
What does this mean for you, the author refreshing your inbox on April 30th? It means that a 5.8 average, which might have squeaked into a poster slot in 2024, now places you firmly in borderline territory. The bar is not shifting because of arbitrary committee decisions; it is rising because the field is maturing. Each cycle brings more polished work from stronger baselines, so the middle of the distribution gets denser and more competitive. If you submitted a paper with a solid result but a narrow scope, the threshold prediction matters less than whether your story connects the dots for reviewers who are reading sixty submissions. A clear narrative beats a marginal score bump every time.
We find the community's focus on the exact number slightly misplaced. The difference between a 6.2 and a 6.4 is noise in the reviewing process, not a signal about your contribution. A more productive approach is to study what made accepted papers from ICML 2025 stand out: they were not the ones with the flashiest architectures, but the ones that made the reviewer's job easy. Clear motivation. Reproducible steps. Honest discussion of limitations. Those qualities do not change with the threshold. They compound.
Set your expectation on the number for planning purposes, then do not give it another minute of worry. Focus your energy on the one variable you actually control: how well your paper communicates its value. The threshold is a mirror of the community's standards, not a judgment on your idea. And on April 30th, regardless of the score, you will have a clearer picture of where your work fits in a field that keeps asking harder questions. That is the real data point worth tracking.