# What does an AI model mean by return on investment?

_We asked 14 AI models which MBA program gives the best return on investment. claude-opus-5 put Stanford first in 9 answers out of 10. gpt-6-astra put it fifth in 7 and left it out of the other 3. Neither model doubted the school. They were doing different sums, and only one of them said which._

Field Notes 05 · 11 Oct 2026 · 4 min read · https://www.brandranking.ai/field-notes/what-does-an-ai-model-mean-by-return-on-investment

---

## The short answer

1. **The question doesn't say what ROI means, so each model decides.** claude-opus-5 weighs the fee against a career of earnings and puts Stanford first. gpt-6-astra counts how fast the fee is paid back and puts INSEAD first, in all 10 answers.
2. **gpt-6-astra tells you its yardstick. claude-opus-5 doesn't.** Every one of gpt-6-astra's #1 picks opens by narrowing the question: "self-funded", "full-time", "faster financial payback over lifetime earnings".
3. **Stanford is the study's #1 and its widest split.** Its brand score of 66 is the highest of any school, but individual models score it anywhere from 0 to 100. Every gpt model scores it 40 or lower; no other model goes below 56.
4. **Fifth isn't a verdict on the school.** Every time gpt-6-astra ranked Stanford, it said financial aid could move it up. Four times it added that the low rank was "not for educational quality".

> **What this measures**
>
> How AI models fill in a word the question leaves open, and what that does to a
> brand's rank.

_The test_

## Same question, two calculations

We asked 14 models "Which MBA program gives you the best return on investment?", 10 times each for the 11 current models and 4 times each for three since retired, answering from training with no web search. Each answer is a ranked top five with a reason for every pick.

Two models read the question in opposite ways:

**Where Stanford landed, 10 answers per model**

| Model | First | Third | Fifth | Not named |
| - | - | - | - | - |
| claude-opus-5 | 9 | 1 | 0 | 0 |
| gpt-6-astra | 0 | 0 | 7 | 3 |

[Open the MBA ROI study](https://www.brandranking.ai/category-ranking/mba-programs-roi)

---

_claude-opus-5_

## The upside outweighs the price

claude-opus-5 doesn't ignore cost. It names INSEAD in all 10 answers and BYU Marriott, the low-tuition pick, in 8. Over half its reasons (58%) mention payback, foregone salary or a one-year format. It just weighs the cost against a career of earnings, and on that sum Stanford comes out on top:

> **claude-opus-5, ranking Stanford first**
>
> "Its network and access to entrepreneurial capital add option value that is
> hard to price."

The one answer where Stanford slipped to third shows the same sum done the other way. BYU Marriott took first because "the payback period is often under two years". Stanford's earnings were still "enormous", but its cost "drags the pure ratio down".

---

_gpt-6-astra_

## The time out costs too much

gpt-6-astra answers a narrower question than the one we asked, and says so up front. Its first answer opens: "Assuming a full-time MBA and prioritizing faster financial payback over lifetime earnings, INSEAD is my top recommendation." The other nine set similar terms, from "for a self-funded applicant" to "weighing tuition, time out of work, and career opportunities rather than salary alone".

Under that lens, program length decides the list:

**Schools gpt-6-astra named, out of 10 answers**

| Name | Context | Share |
| - | - | - |
| INSEAD | 10 months | 10 |
| Cambridge Judge | 1 year | 10 |
| Oxford Saïd | 1 year | 9 |
| Stanford | 2 years | 7 |
| Kellogg | For its one-year option | 5 |
| HEC Paris, IMD, ISB, IIM Ahmedabad, SDA Bocconi | 1 year to 16 months | 1–2 each |

_Harvard and Wharton, named by 13 and 12 of the 14 models, never appear._

**10 of 10** gpt-6-astra answers that put INSEAD first

Every reason for INSEAD comes back to its short format: "Its roughly 10-month format limits foregone salary". Stanford is the only two-year program to make the list at all, and it comes with a caveat attached:

> **gpt-6-astra, ranking Stanford fifth**
>
> "It ranks fifth under this faster-payback lens—not for educational quality—and
> could rank first for someone with substantial financial aid or a longer
> investment horizon."

All 7 of gpt-6-astra's answers that rank Stanford say financial aid could move it higher.

---

_The panel_

## The other models take sides

The two readings run through the whole panel. Two first picks in three went to a big-name two-year school, and one in three to a one-year or low-tuition program.

| Source | Share |
| - | - |
| Stanford | 48% |
| INSEAD | 21% |
| Wharton | 16% |
| Everything else | 15% |

_First picks in the 120 answers that named one. Everything else is mostly BYU Marriott (9%) and Chicago Booth (3%)._

In the [watch study](https://www.brandranking.ai/field-notes/how-consistent-are-ai-models-when-ranking-brands), two models from one lab could rank very differently. This time the lens runs in the family. gpt-5.6-luna put BYU Marriott first in half its answers, and the four gpt models are the four lowest scores for Stanford.

**Stanford's brand score, by model**

| Name | Share |
| - | - |
| gemini-3.5-flash-lite | 100 |
| claude-opus-5 | 96 |
| kimi-k3 | 96 |
| grok-4.6 | 94 |
| gemini-3.8-flash | 84 |
| qwen3.8-max | 84 |
| mistral-medium-3.5 | 68 |
| deepseek-v4.1-flash | 58 |
| claude-haiku-4.5 | 56 |
| gpt-5.6-luna | 34 |
| gpt-6-astra | 14 |

_Three retired models answered 4 times each: gemini-3.7-flash scored Stanford 100, gpt-5.6-sol 40 and gpt-5.6-terra 0. The study's 66 averages all 14._

The payback lens isn't only OpenAI's, though. gemini-3.8-flash put INSEAD first in 8 answers out of 10, with Stanford second.

---

_What it means_

## Find out which sum the model is doing

In [Note 03](https://www.brandranking.ai/field-notes/does-a-more-precise-question-give-a-different-brand-recommendation), adding a word to the question changed nothing. Here the model narrowed the question by itself, and that moved Stanford from first to fifth.

_01_

**A vague word in the question gets filled in by the model.** "Best", "value" and "ROI" each hide a yardstick, and each model picks its own. Before reading your rank, read the reasons: they tell you which yardstick you were measured on.

_02_

**A low rank can come with its own counter-argument.** gpt-6-astra never doubted Stanford's quality. It named the one fact that would change its answer, financial aid. For a brand that loses on one yardstick, that caveat is the fact worth making easy to find.

> **The question behind the ranking**
>
> When you say ROI, what counts? Lifetime upside, or faster payback?

**Which yardstick are the models using on your brand?** Every BrandRanking.AI study asks a panel of models the same question about 10 times each and keeps every reason they give. Browse the [category studies](https://www.brandranking.ai/categories) or request a study for your own market.

- [Open the MBA ROI study](https://www.brandranking.ai/category-ranking/mba-programs-roi)
- [Request a study](https://www.brandranking.ai/#request-category-study)

---

## How we measured

- **Question.** "Which MBA program gives you the best return on investment?", run on 4 and 23 September 2026, 10 times per model and 4 times per retired model. 122 answers came back; two from deepseek-v4.1-flash were empty.
- **Models.** claude-opus-5, claude-haiku-4.5, gpt-6-astra, gpt-5.6-luna, gemini-3.8-flash, gemini-3.5-flash-lite, grok-4.6, deepseek-v4.1-flash, kimi-k3, qwen3.8-max and mistral-medium-3.5, plus three since-retired models with 4 answers each: gemini-3.7-flash, gpt-5.6-sol and gpt-5.6-terra.
- **Scores.** A brand gets 100 points for a #1, 80 for a #2 and so on down to 20 for a #5, and 0 when it's left out. The points are averaged for each model, then across models with every model counted equally.
- **Payback language.** A reason counts if it mentions a one-year or 10-month format, opportunity cost, foregone or lost earnings, payback, breakeven or time away from work.
- **What this is not.** These are model recommendations, not measured returns on an MBA. Program lengths are each school's standard full-time format.

_BrandRanking.AI © 2026 · Study data: [brandranking.ai/categories](https://www.brandranking.ai/categories)_
