Thirty highly selective American colleges published their 2024-25 Common Data Sets this winter. Read together, the numbers describe a market that has tightened on every margin at once: rates lower, applications higher, binding rounds dominant, and tests on the way back in.
Once a curiosity, the sub-4% club is now crowded. For the Class of 2029, Caltech (2.6%), Vanderbilt RD (3.3%), Stanford (3.6%), Harvard (3.7%), UPenn RD (3.7%), and Yale (3.9%) all admitted fewer than four applicants in a hundred.
These are overall rates. Strip out the binding early round — where athletes, legacies, and committed applicants stack — and the regular pool gets a steeper drop. At Vanderbilt, RD acceptance has effectively converged on Harvard's headline number.
If RD is unforgiving, what does the binding round buy?
Binding Early Decision lets a student trade optionality for a statistical edge. Across the schools that publish split rates, that edge is large — 2x to 6x the regular round, depending on the institution.
Middlebury sits at the extreme: a roughly 30% combined ED rate against a 5% RD rate — a 6x multiplier — with 68% of the freshman class filling via binding rounds. WashU (26%), Northwestern (25%), and Williams (25%) cluster close behind.
Cornell (1.6x) and CMU (1.5x) anchor the other end. The headline ED rate is partly a composition effect: hooked applicants (athletes, legacies, recruited artists) disproportionately apply ED, so the unhooked-applicant boost is smaller than the raw ratio.
Both rounds, though, sit on top of an applicant pool that keeps growing.
UCLA received roughly 146,000 freshman applications for Class of 2029. Michigan crossed 109,000. UVA, Cornell, UPenn, Columbia, and Yale all moved more than 50,000 files. The mechanical effect: even with growing class sizes, denominators run away from numerators.
Read against admit counts, the divergence becomes visceral. Caltech reviewed 13,856 files to admit 356. UCLA reviewed 146,000 to admit roughly 13,000. Both arrive at single-digit rates by very different routes — one extreme selectivity, the other extreme volume.
Numbers tighten. So do the academic profiles inside them.
Among schools that report scores, the middle 50% lives in a narrow vertical strip: 1480 to 1580. Harvard and MIT top the chart at 1510-1580 and 1520-1580 respectively; Yale, Columbia, Brown, and Vanderbilt cluster around 1500-1560.
Two cautions. Test-optional self-selection inflates the band: weaker test-takers increasingly do not submit. And the tied medians around 1540-1550 do not mean these schools admit interchangeable applicants — the score is a floor, not a sorter.
UCLA reports no band. The University of California system is fully test-free; SAT/ACT scores are not considered. Michigan and UVA have the widest reported ranges (1350-1530 and 1410-1520), reflecting larger, more academically diverse classes.
Once admitted, who actually shows up?
Yield — the share of admitted students who enroll — is the metric that keeps deans awake. Three clusters emerge clearly.
At the top, an elite tier above 80% is essentially MIT (86%), Harvard (84%), and Stanford (82%) — the schools that win nearly every cross-admit decision. Princeton (78%) and UChicago (~80%) sit just below, the latter held up by aggressive ED use.
A strong middle from 60-75% catches most of the Ivy League: Cornell, UPenn, Yale, Brown, Dartmouth all sit there. And a lower tail around 35-45% contains schools like Tufts (~35%), UVA (~38%), Rice (~38%), and Amherst (~39%) — institutions that admit deep into a national pool but compete with peers in the cross-admit round.
Yield drives strategy. So does the size of the entering class.
Selectivity is partly a function of capacity. Caltech enrolls ~235 freshmen a year. Amherst seats 480, Williams 560, Middlebury 600. At the other end: Michigan enrolls ~7,200, UCLA ~6,500, UVA ~3,800.
Mid-sized privates — Yale, Princeton, Harvard, Brown — cluster between 1,400 and 1,750. The Ivies are not large universities; they are mid-sized colleges that read enormous application piles. UPenn (~2,400) and Cornell (~3,600) are the outliers among the Ivy Eight.
A footnote that became a headline: testing came back.
In 2020, the entire Ivy League went test-optional. The phrase "test-optional era" was treated like a fixture. By 2024-25 it was clearly transitional.
MIT reinstated the SAT/ACT requirement in 2022. Dartmouth, Brown, Yale, Harvard, Stanford, and Caltech followed for the Class of 2029 (Fall 2025) cycle. UPenn and Cornell are returning for 2026. Among this set, only UChicago, Columbia, JHU, Duke, Northwestern, Notre Dame, and most LACs remain test-optional.
UCLA and UC Berkeley are the policy outliers in the other direction: they remain permanently test-blind, refusing to consider scores even when submitted.
For one slice of this group, geography also rewrites the math.
Three schools in this set are flagship publics: UVA, Michigan, and UCLA. Their admit math splits sharply by residency.
At UVA, in-state acceptance runs around 26%; out-of-state, roughly 14%. At Michigan, in-state ~39% against OOS ~18%. UCLA is the exception: California residents are prioritized through enrollment targets, not by rate — in-state and out-of-state admit rates run comparably, both around 9-11%.
For an out-of-state applicant, a "public Ivy" is no easier than a private peer. For an in-state one, it can be the most efficient ticket in the country.
Each of the eight charts above is a different cross-section of the same admissions market. Cumulatively, they describe a system that has compressed on every dimension at once: lower rates, higher volumes, narrower score bands, and a bigger share of class committed before spring even arrives.
Strategy follows arithmetic. ED is no longer a niche bet; at schools like Middlebury, WashU, and BC, it is the modal route into the freshman class. Testing has reversed in the opposite direction at the top tier — eight schools in this set are back to required, or will be by Fall 2026. And out-of-state applicants to flagship publics are competing in what is effectively a different admissions cycle than residents.
For a single applicant, the practical takeaway is mundane and important: the distributions matter more than the headline. A 4% overall rate hides a 25% ED rate and a 3% RD rate; a 9% UCLA rate hides a math problem with no SAT in it. Read the splits.