At a glance
MIT admits around 4.6 percent of applicants, yet computing there is now several majors rather than one. A look at the enrolment data, the AI shift and what other departments can learn from it.
MIT has admitted between roughly 4.5 and 4.8 percent of its applicants for four consecutive cycles. For the Class of 2030, 28,349 students applied and 1,299 were offered places; the year before, 29,281 applied and 1,334 got in. The admit count barely moves. What shifts is where the students inside that number end up once they arrive, and that is where the story about MIT computer science actually lives.
From outside, the reputation looks flat and monolithic: the best computing department in the world, full stop. QS ranked MIT first in computer science and information systems for a 15th consecutive year in its 2026 subject tables, with an overall score of 94.9 against Stanford's 92.9 and Carnegie Mellon's 90.9. Inside the Institute, the department is not one thing. It's a group of degrees with different mathematics cores, different prerequisites and, increasingly, different kinds of students.
What Course 6 actually contains
MIT numbers its departments rather than naming them in the catalogue. Electrical engineering and computer science is Course 6, taught by the Department of Electrical Engineering and Computer Science. Since 2019 that department has shared the computing brief with the Schwarzman College of Computing, which houses the artificial intelligence programmes and the Institute-wide computing units.
A student inside Course 6 still has to pick a specific degree. The oldest and best known is 6-3, Computer Science and Engineering, which is what most people mean when they say MIT computer science. The newest large one, 6-4 Artificial Intelligence and Decision Making, took its first students in fall 2022. There are joint programmes with biology, brain and cognitive science, economics, urban planning and mathematics, plus a newly renamed electrical engineering with computing degree, 6-5, that drew 94 first-year declarations in a single spring and became the third-largest major at the Institute.
| Course | Degree | Total majors, 2026–27 |
|---|---|---|
| 6-3 | Computer Science and Engineering | 544 |
| 6-4 | Artificial Intelligence and Decision Making | 354 |
| 6-5 | Electrical Engineering with Computing | 249 |
| XVIII-C | Mathematics with Computer Science | 88 |
| 6-2 | Electrical Engineering and Computer Science | 85 |
| 6-9 | Computation and Cognition | 43 |
| 6-14 | Computer Science, Economics and Data Science | 39.5 |
| 6-7 | Computer Science and Molecular Biology | 32 |
Read that table with two cautions. The registrar counts joint majors as half a student in each department, folds in secondary majors, and the department's own annual reports tally slightly differently. Across recent published counts, 6-3 has landed somewhere between roughly 540 and 720, and 6-4 between roughly 330 and 370. The direction of travel is the signal; the decimal place is noise.
The growth went to the AI major
Course 6-3 grew 16.4 percent between 2016 and 2022, from 707 students to a peak of 823. Then it turned. By 2025 it stood at 672, an 18.3 percent fall, and the department's report to the president recorded 715 core 6-3 majors in academic year 2025 against 804 the year before and 831 the year before that. Over the same window 6-4 went from 37 students to 372. The student newspaper's analysis of registrar data tracks the two curves crossing and notes the obvious candidate explanation: the newer major absorbed students who would otherwise have declared the older one.
The curriculum backs that up. 6-3 requires Software Construction; 6-4 doesn't, weighting probability and statistics instead. For a student who wants to work on models and inference rather than build large systems, the path is shorter and the maths is more central.
Course 6 in total still grew 21 percent from 2016 to 2025 despite all this shuffling. But it fell 3.5 percent between 2024 and 2025, from 1,672 students to 1,614 — the first annual decline in a decade. Asu Ozdaglar, deputy dean of academics at the Schwarzman College of Computing, told the New York Times that students who prefer working with data to solve problems gravitate to the AI major, and that applicants interested in applying it in biology and health care are arriving too. That's a reallocation, not a retreat.
Photo by Poddar Group of Institutions on Unsplash
Research access is the product
The reason the queue forms is less mysterious than the branding suggests. MIT's Undergraduate Research Opportunities Program, started in 1969 as one of the first of its kind in the United States, is the mechanism. In the 2024–25 year more than 3,200 undergraduates took part, 58 percent of the faculty served as mentors, and 93 percent of the Class of 2025 completed at least one project before graduating. Nearly 77 percent of projects were paid rather than credit-bearing, and the Institute put $16.8 million behind undergraduate research support. The program's own figures also show 73 percent of first-years participating, which matters: the pipeline is opened in year one, not senior year.
Scale helps in other ways. MIT runs about 4,500 undergraduates against a reported 3-to-1 student-faculty ratio, and its introductory programming course through MITx has registered more than 140,000 learners in a single offering according to the department's 2022 annual report. The on-ramp is global. The degree is not.
Then there is the alumni network, which is where the numbers get big enough to need scepticism. A 2015 study by Edward Roberts, Fiona Murray and Daniel Kim, working from a 2014 survey of MIT's then 104,000 living alumni, estimated that alumni had founded 30,200 active companies employing about 4.6 million people and generating around $1.9 trillion in annual revenue. Those are all-alumni figures spanning every discipline and a decade old, so treat them as evidence of a culture rather than a forecast for a computer science graduate in 2026. The relevant part is the mechanism: 25 percent of surveyed alumni had founded a company, and roughly 40 percent of those were serial founders.
The queue has a price
Popularity at this level creates a narrow gate. The Institute's published admissions statistics for the Class of 2029 show 136 international applicants admitted out of 6,926, a rate of 2.0 percent against 4.6 percent overall. Early Action admitted 5.98 percent that cycle while Regular Action admitted 3.50 percent. For the Class of 2030, Early Action drew 11,883 applicants for 655 offers, and 7,738 of those early applicants were deferred into the regular pool rather than decided either way.
Testing is required, and the middle 50 percent of admitted students sits at an SAT composite of 1520 to 1570 with maths scores clustered at 780 to 800. MIT has been explicit that it doesn't reward demonstrated interest, which means campus visits and family ties buy nothing. Financial aid is the one lever that widens the funnel: since the 2025–26 year, families with typical assets earning under $200,000 pay no tuition, and those under $100,000 have no expected parental contribution at all. Applications still slipped slightly for the Class of 2030, so don't assume the policy has reset the pool yet.
Look outside Cambridge and the picture gets more complicated. The Computing Research Association counted nearly 173,000 undergraduates in computing programmes in spring 2024, up from about 65,000 a decade earlier, but reporting by the New York Times in December 2025 found 62 percent of computing programmes recording enrolment declines that autumn, with job-market anxiety the cited driver. MIT's computing numbers have held because its applicant pool is global and its brand is unusually durable. Most departments are not in that position, and shouldn't plan as if they were.
What this means for your lab
A colleague — call her Dr. M — chairs computing at a mid-sized public university and spent last spring rebuilding a curriculum around a declining major. Her department had split its single computing degree into a general track and a data-focused track, hired two teaching-stream staff against projected growth, and then watched total declarations fall while the split held roughly steady. She now describes the reorganisation as expensive relabelling. One department's experience isn't a trend, but the arithmetic behind it is common: splitting a major rarely grows the major, it redistributes it.
The MIT data offers three usable lessons for anyone planning a computing programme, and one uncomfortable one.
- Read total enrolment, not the headline major. Course 6 lost its decade-long growth streak the same year its AI degree hit record numbers. A single named programme can look healthy while the department shrinks.
- Keep the mathematical core shared. MIT's 6-4 didn't abandon rigour when it dropped Software Construction; it swapped in more probability and statistics. Differentiation that lowers the mathematical floor tends to produce graduates employers discount.
- Build the undergraduate research pipeline in year one. The 73 percent first-year participation rate is the number to copy, not the 93 percent graduation figure, because the second is a consequence of the first.
The uncomfortable lesson is about capacity. MIT can absorb a shift of several hundred students between majors because it has the teaching depth to staff both. A department running at the edge of its teaching budget cannot, and redesigning around whichever specialism is trending in a given admissions cycle is how programmes end up with unfilled faculty lines and over-subscribed service courses in the same year.
Photo by Ahmad Faiz on Unsplash
One thing to do before your next curriculum vote
Ask your registrar for three years of major-declaration counts broken out by the specific degree, plus enrolment in your gateway programming sequence, and compare the two lines on one page. If declarations are flat and the gateway course is shrinking, you have a demand problem upstream and a new major won't fix it. If declarations in a new track are growing while the total is flat, you have relabelling, and the honest question is whether the new label serves students or serves the prospectus. Pull the numbers before the committee meets, not after the hiring plan is written.
- MIT computer science
- Course 6
- AI major
- computing enrollment
- higher education

