GMAC published the August 2025 percentile table yesterday. We closed the 2024 brief — the one written for last year's much larger refresh — with a forecast: "the next one will be in roughly twelve months, and barring a structural change to the test itself, it will move a smaller distance than this one did." The forecast held. The 2025 refresh is a small shift, concentrated at the same end of the curve, in the same direction. This is what changed and what we updated.
The headline shift
A 705 now reads as the 97.8th percentile, against last year's 98.2nd. The 99th percentile threshold sits at a 755, against last year's 745 and the early-2024 705. The middle of the curve barely moved — scores between 555 and 645 are inside half a percentile point of their August 2024 positions. The lower bands are unchanged within rounding. The entire shift, once again, is above 695.
The direction is the same as last year, but the magnitude is roughly half. This is what cohort stabilisation looks like — the rolling five-year window now contains two full years of Focus-Edition data, and the marginal effect of adding another year is smaller than the effect of adding the first full year was. Future refreshes should keep getting smaller until either GMAC changes the test's structure or the candidate population shifts in a new direction.
Why this happened — same answer, less of it
The 2024 brief named two forces. Both are still operating, both with less amplitude.
First, the five-year rolling window. The 2024 refresh was the first table that included a full year of Focus data; the proportion of Focus responses in the window roughly doubled compared to 2023. The 2025 refresh added one more year of Focus data to a window that already had one — a marginal change, not a structural one.
Second, the population shape. The composition shift — Verbal slightly weaker on average, Quant materially stronger — kept the same sign but a smaller magnitude. Verbal scaled means moved by roughly a tenth of a point; Quant scaled means moved by roughly four-tenths. The composite shift remains a Quant-driven shift, but the size of the Quant move is decelerating.
What this means for target scores
For percentile-anchored targets, the same adjustment as last year still applies, in a smaller dose. A candidate who in early 2024 was targeting "the 95th percentile" adjusted their scaled-score target up by 10 to 20 points last August. This August, that same candidate's target moves up by another 5 to 10 points against the new figures.
The 2024 brief noted that the "M7-competitive" floor was being quietly raised by application consultants from the 705 of early 2024 toward 715–725. The 2025 consensus, as best we read it, has the floor sitting at 715–725 still, with the upper end of that band now sitting roughly where 735–745 used to.
For scaled-score targets, the answer is unchanged from last year. The score itself has not gotten harder to obtain. The item parameters are the same. What has changed is the company that 705 puts the candidate in.
What we updated
Three things changed on the Brightroom surface this morning, ahead of the table going live.
The predicted-band model. Every diagnostic, every Mock Exam, every Section Analytics readout that surfaces a percentile alongside a scaled score reads from the August 2025 table starting today. Older sessions are re-percentile-coded on read, not on write — so historical scaled scores remain stable and the percentile column updates everywhere it appears, in one operation.
The target-score recommender. Candidates whose target is set against a percentile benchmark (top 5%, top 1%, etc.) get the recommendation against the current table. We do not retroactively move a candidate's scaled-score target — that is a contract — but we do surface a small note when the percentile equivalent of their target has shifted by more than half a band. About 6% of active candidates get that note this week.
The Ultra guarantee floor is unchanged. The 715+ absolute floor for diagnostics ≥ 595 is written in scaled-score points, not in percentile ranks, precisely because percentiles drift and scaled scores do not. An Ultra candidate who was promised 715+ is still promised 715+ — what the percentile equivalent of 715 means in 2025 is a different number than it meant in 2024, and that is the candidate's choice, not the contract's.
What we did not change
We did not re-calibrate item parameters. The (a, b, c) values on every item in the pool are estimated from candidate response data, not from population percentiles. A change in how scaled scores map onto percentiles is downstream of item calibration, not upstream of it.
We continue to note this because at least one competing platform has, again, published a "we have re-calibrated our adaptive engine in response to the percentile change" claim this week. That sentence still does not describe a coherent psychometric operation. The percentile column on the report should change. The engine should not.
What to do
Two adjustments. Same as last August, less of them.
Re-read the target. If a candidate set a scaled-score target against percentile-anchored advice published before this morning, the percentile equivalent of that target has moved — by a smaller amount than last year, but moved. Adjust the scaled-score target up by 5 to 10 points if the percentile match matters. Leave it alone if the scaled score was always the target.
Do not chase the rebase. A week of Quant grinding to "recover" a tenth of a percentile point is the wrong response. The marginal cost of pulling a Quant scaled score from 82 to 83 has not changed. What has changed is what 82 means against the population. Train the skill, not the statistic.
Closing
GMAC publishes an updated table every August. We will publish a brief like this each time. The next one will be in roughly twelve months, and we expect it to move less than this one did — for the same reason this one moved less than the 2024 refresh did.
— Brightroom Research