Machine Learning Conference Acceptance Rates: NeurIPS, ICML, ICLR, CVPR, AAAI

If you are deciding where to send your next paper, the acceptance rate is one of the first numbers you look up. It is a rough measure of how selective a venue is, and at the top machine learning conferences it has become a closely watched statistic as submission volumes climb year after year. This post collects the most recent acceptance rates for the five biggest AI and machine learning conferences, NeurIPS, ICML, ICLR, CVPR, and AAAI, and looks at what the trend lines actually say.

Why acceptance rates matter (and what they miss)

An acceptance rate is simply the number of accepted papers divided by the number of submissions in a given cycle. A lower rate signals a more competitive venue, which usually means more submissions chasing a limited number of program slots. It is a useful yardstick, but it is not the whole story. A venue can post a higher rate because it runs multiple tracks or accepts short papers, and a lower rate can reflect an overloaded review process rather than higher quality. Reviewer scores, the reputation of the program committee, and the community a conference serves all matter at least as much as the headline percentage.

With that caveat in mind, here are the numbers.

The numbers: 2025 acceptance rates

All figures below are for the main technical track, full papers, from the most recent completed cycle (2025).

NeurIPS

The Conference on Neural Information Processing Systems is the largest venue in the field and, by submission count, the largest academic conference in machine learning anywhere. The 2025 main track drew 21,575 valid submissions and accepted 5,290, an acceptance rate of 24.5%. That is roughly a 60% jump in submissions over the previous year, yet the acceptance rate held steady in its usual mid-twenties band, which means the program expanded capacity rather than tightening the bar. NeurIPS has sat around 24 to 26% for several years running. More on the conference: NeurIPS.

ICML

The International Conference on Machine Learning is the second of the three core ML venues. ICML 2025 accepted 3,260 papers from 12,107 submissions, an acceptance rate of 26.9%. That is a slight tightening from 2024, when it accepted 27.5% from 9,473 submissions, so the rate edged down even as submissions grew by more than a quarter. ICML remains the most consistent of the big three, rarely straying far from the mid-to-high twenties. See ICML.

ICLR

The International Conference on Learning Representations is the youngest of the three core venues and runs a fully open review process on OpenReview. ICLR 2025 reviewed 11,565 submissions and accepted about 32.1% (roughly 3,704 papers, of which 213 were orals and 380 spotlights). ICLR has long carried the highest acceptance rate of the top ML conferences, generally in the low thirties, which makes it a relatively attractive target if your work fits the representation-learning scope. Details on ICLR.

CVPR

The Conference on Computer Vision and Pattern Recognition is the flagship for computer vision and, after NeurIPS, one of the largest by submission volume. CVPR 2025 received 13,008 valid submissions and accepted 2,878, an acceptance rate of 22.1%. That is down from 23.6% in 2024, continuing a slow decline as the field grows. CVPR is consistently the most selective of the five conferences here, hovering in the low twenties. See CVPR.

AAAI

The AAAI Conference on Artificial Intelligence is the broadest of the five, covering planning, knowledge representation, robotics, and reasoning alongside machine learning. AAAI 2025 set a record with 12,957 submissions and accepted 3,032, an acceptance rate of 23.4%, essentially flat against 2024’s 23.7% despite more than 3,000 additional submissions. AAAI’s main track has settled into the low-to-mid twenties. More on AAAI.

Trends and takeaways

A few patterns hold across all five venues:

• Submissions are exploding, acceptance rates are not. Every conference here saw double-digit growth in submissions, in NeurIPS’s case roughly 60% in a single year. Yet acceptance rates barely moved. Program committees are scaling the number of accepted papers to keep the percentage roughly constant, rather than holding the headcount fixed and letting the rate collapse.
• The real bar is higher than the rate suggests. Because submission quality is rising and reviewer load is heavy, a borderline paper at a 25% venue can still be rejected with respectable scores. Several 2025 cycles saw well-reviewed papers rejected simply because the marginal slot got more competitive than the percentage implies.
• The venues are clustered tighter than people think. Excluding ICLR’s slightly higher band, the rest sit between 22 and 27%. The choice between them is usually about fit and community, not about gaming a lower rate.

Conclusion

For 2025, the headline machine learning acceptance rates were NeurIPS at 24.5%, ICML at 26.9%, ICLR at 32.1%, CVPR at 22.1%, and AAAI at 23.4%. All five remain highly competitive, and all five are getting more submissions every year. Use the rate as a starting reference, but choose where to submit based on scope, the community you want to reach, and the kind of impact you want your work to have. You can browse dates, locations, venues, and hotel guides for each of these on their conference pages: NeurIPS, ICML, ICLR, CVPR, and AAAI.