What is the best medical physics program? I asked eight LLMs
Published:
Everyone in this field gets asked which program is best, and everyone has an opinion. Nobody has a ranking. US News ranks law schools, business schools, hospitals and engineering departments, and skips graduate medical physics entirely. So I got curious and asked the machines.
Conflict of interest, stated up front: I am a PhD student at Wisconsin, which finished first. Weigh everything below accordingly.
Rankings get made by people who pick the criteria, weight them, and publish the result. Medical physics never got that treatment. It is a small field, under a hundred accredited graduate programs in the United States, and no major outlet bothers to rank them. Applicants compare them by asking around, which is how I ended up with opinions of my own.
Increasingly, applicants ask a chatbot instead. I wanted to know what it would tell them. So I gave eight large language models, one from each major company, the same question in a fresh conversation with no memory of the last one:
Rank the top 15 graduate programs in medical physics in the United States.
Give a numbered list from 1 to 15 containing only the institution name. Do not provide explanations, commentary, caveats, or additional text.
Reworded AWhich 15 graduate programs in medical physics are the best in the United States?
Reworded BList the 15 leading U.S. graduate programs in medical physics in rank order.
Each model got that question ten times, which is eighty answers to that wording alone. Two rewordings of it, plus a version naming master’s programs and one naming doctoral programs, brought the total to 320. The calls went out one at a time, and I logged every one with a hash of the response and a note of which company’s servers had actually answered it.
Here is the field. One model per company, pinned to an exact version and serving provider so the same weights answered every time:
| Model | Company | Exact version |
|---|
| Model | Endpoint | Temperature | Max tokens | Reasoning |
|---|
Two of the eight will not accept a temperature setting at all, and four cannot switch their reasoning off, so those run at the lowest effort their provider allows. Holding every model to identical settings is impossible, so the table records what each one actually accepted.
What comes back is a ranking. A confident one, with clear favourites and a sensible-looking tail. The only question worth asking is what it is made of.
The consensus top 15
Pooled across 8 models, 10 answers each. Every model gets one vote.
Show the numbers
The list holds up better than it should
Thirteen of the fifteen hold CAMPEP accreditation, which is the field’s actual quality bar and the closest thing to an answer key. Wisconsin, MD Anderson and Duke lead, and any medical physicist would put those three near the top of their own list. For a ranking assembled by next-token prediction, that is a respectable showing.
Two do not belong. Michigan lands fifth and Harvard tenth, and neither appears on the CAMPEP graduate program list. Michigan showed up in 58 of 80 answers, so the models have settled on it.
Across every answer the models named 59 distinct schools, and 35 of those hold no accredited graduate program in the field. Most appear once or twice. The models know the top of the field well and improvise below it, which is roughly what a well-read undergraduate would do.
Master’s or doctoral? The models shrug
All fifteen for each phrasing. Rank sits at the left of each row.
One thing the colours cannot carry: they mark schools holding an accredited graduate program in medical physics at any degree level, because the graduate list is what I checked against. Since CAMPEP accredits master’s and doctoral programs separately, a blue bar in the master’s column is not a promise that the school runs an accredited master’s — and some of the schools the models name here do not offer that degree at all. That is arguably the sharper version of the finding: asked for the best master’s programs, the models return departments that have no master’s program to rank.
Show the numbers
So where is it getting this?
Across the 79 graduate answers the models named 35 schools that hold no CAMPEP-accredited graduate program. That is a narrower claim than it sounds. CAMPEP accredits graduate programs, residencies and certificate programs on separate lists, and several of these schools run an accredited residency or certificate program. What none of them runs is an accredited graduate program, which is what the prompt asked for.
Look at the list and the pattern is obvious. These are famous universities. The models appear to be reaching for general institutional reputation and letting it stand in for a field they have little specific information about, which is roughly what a well-read undergraduate would do when put on the spot.
Schools with no accredited GRADUATE program in medical physics
Times named across the 79 graduate answers. Darker segment counts top-ten finishes.
Show the numbers
Rephrasing the question barely moves the needle
Each answer hands out ten top-ten places. The share of them that went to a school running no accredited graduate program.
Show the numbers
Do they agree with each other?
At the top, strongly. Wisconsin, MD Anderson and Duke clear nine of ten from almost every model. Past Chicago the grid breaks apart, and several schools survive on one model’s enthusiasm. Gemini alone carries Johns Hopkins. DeepSeek alone carries UC Berkeley/UCSF.
Who picked whom
How many of each model’s 10 answers put the school in its top ten.
Show the numbers
One model refused to play
Grok answered the graduate question ten times. Nine times it produced a tidy list of fifteen schools. On the ninth attempt, same prompt and same settings, it stopped and said this:
Grok’s ten answers to the graduate question, in order.
It declined twice more on master’s and twice again on doctoral. Claude did something similar once on a reworded question, explaining at length that no such ranking exists in a form it could reproduce.
Grok was right, which is the awkward part. There is no official ranking. Every list on this page is fabricated in exactly the sense it meant. It just happened to be outvoted nine to one by its own other answers.
So what is this worth?
About as much as any ranking, which is the honest answer. Every ranking you have read was somebody choosing criteria and weights. This one is eight language models averaging whatever they absorbed about American universities, with no criteria at all. The exact prompt, the exact model versions, the exact date and the raw counts are all here, so you can see how the sausage got made. Most rankings do not offer that.
The reason it matters is that this is plausibly how a lot of applicants will shortlist programs from here on. Not by reading accreditation lists. By asking a chatbot, once, and taking the answer. Anyone who does that in medical physics gets Michigan fifth and Harvard tenth.
Which would sting more if program rank were the thing that mattered. It mostly is not — and “graduate” is doing a lot of quiet work in that sentence, because it covers two degrees that get decided on completely different grounds.
For a PhD, the department barely registers next to the person supervising it: what they research, how they mentor, whether they have funding that will still be there in year four, whether you can stand working with them for five years. Two students in the same accredited department can have completely different educations depending on whose lab they join. No ranking captures that, and no language model can tell you who your advisor should be.
A master’s is a different question, and a harder one to put to a chatbot. It is a short professional degree pointed at a residency, so the things that decide whether it was worth it are mostly ones a ranking never sees. What does it cost, and is any of it funded or offset by a stipend? How much clinical time do you actually get, in whose clinic, and on whose machines? And above all, where did recent cohorts end up — because in this field the residency match is the bottleneck, not the degree. Most accredited programs publish outcome data somewhere on their own site: completion rates, and where their graduates went. Go read it. A program whose graduates consistently match into residencies is telling you something a top-15 list cannot, and a program that is vague about it is telling you something too.
So ask the chatbot if you like. Then go read the faculty pages, email a few of them, talk to their current students, and look up the placement numbers. That is the part that decides your degree.
And yes, Wisconsin came first, and yes, I go here. I would have told you it belonged near the top before I ran any of this. That is exactly the problem with letting anyone, or anything, hand you a ranking.
320 responses collected 14 September 2026 through OpenRouter, one call at a time, with the model and serving provider verified on every call. Total cost, $1.28. Accreditation checked against the CAMPEP GRADUATE program list as published on the collection date, 62 programs. CAMPEP accredits residency and certificate programs on separate lists, which were not consulted, so “no accredited graduate program” never means a school has no accredited program of any kind. School names were reconciled by hand, because automated matching proposed merging the University of Iowa with the University of Miami.