# TMUA sittings compared: October 2025 against January 2026

> TMUA candidates averaged 4.62 in October 2025 and 3.78 in January 2026. One set of scaling constants covers both, so the gap is composition, not difficulty.

[Formatted version of this page](https://www.frontiervue.com/exams/tmua/guides/sitting-comparison)

Choosing between the October and the January sitting is the first real decision a TMUA candidate makes, and it usually gets made on hearsay. UAT-UK's 2025/26 technical report prints the score distribution separately for each event, and the annual report adds each sitting's candidate numbers and its change on the year. Together they answer three questions precisely: what each sitting's scores looked like, how the scale was built, and where a given score falls inside each group. This page will not pick a sitting for you. It lays out the printed figures, and it spends most of its length on the step readers most often get backwards.

## Key takeaways
- October 2025: 9,685 candidates, mean 4.62, median 4.5, 90th percentile 7.0. January 2026: 6,407 candidates, mean 3.78, median 3.6, 90th percentile 5.9.
- Both events used the same scaling constants, 4.6046 and 1.6369. The report says this was done so that scaled scores stay comparable across events, which is why the gap in means is not a difficulty gap.
- October's median of 4.5 and 90th percentile of 7.0 are not coincidences: the scale was pinned to the October ability distribution, at theta values of -0.0639 and 1.4634.
- The report gives its own reason for the volume difference: candidates may sit only once per cycle, and those applying to the University of Cambridge must sit in October, which accounts for the higher October volume.
- The same score sits at a different percentile in each event, but means the same thing: the report states that a candidate scoring 6.5 in either month has higher ability than one scoring 4.2 in either event.

### Quick answer: October against January

| Item | Value |
| --- | --- |
| Candidates per sitting | 9,685 in October 2025 and 6,407 in January 2026, 16,092 across the cycle |
| Where each sitting centred | October mean 4.62 and median 4.5; January mean 3.78 and median 3.6 |
| One scale or two | One. Both events used the constants 4.6046 and 1.6369 on the same 1.0 to 9.0 scale |
| Can you pick a softer sitting | No. The ruler did not move, so the gap reflects who sat the test, not an easier sitting |

> This page uses only figures printed as text in UAT-UK's reports. Nothing is read off a chart image and nothing is derived across tables.

## Contents
1. The two sittings, side by side
2. Why the gap between the sittings is not a difficulty gap
3. Who sits in October, who sits in January
4. The same score, a different position in each group
5. What these numbers cannot settle, and how to decide

## 1. The two sittings, side by side

The TMUA ran twice in the 2026 admissions cycle: October 2025 and January 2026, 16,092 candidates in total. Table 4 of the technical report prints the scaled score distribution separately for each event, and Tables 1 to 3 of the annual report give each sitting's bookings, delivered tests, share of the sitting, year-on-year change and geography. Here are those columns next to each other, before any interpretation.

- **9,685 / 6,407** Candidates sitting: October / January
- **4.62 / 3.78** Mean scaled score: October / January
- **4.5 / 3.6** Median, the 50th percentile: October / January
- **7.0 / 5.9** 90th percentile: October / January

|  | October 2025 | January 2026 | Whole cycle |
| --- | --- | --- | --- |
| Tests delivered | 9,685 | 6,407 | 16,092 |
| Bookings made | 10,109 | 6,780 | 16,888 |
| TMUA share of that sitting's bookings | 46.7% | 52.9% | 48.9% of all tests taken across the cycle |
| Change from 2024-25 | +8% | +32% | Not printed for TMUA at cycle level |
| Mean scaled score (SD) | 4.62 (1.77) | 3.78 (1.57) | 4.28 (1.74) |
| 25th / 50th / 75th percentile | 3.3 / 4.5 / 5.8 | 2.7 / 3.6 / 4.6 | 3.0 / 4.1 / 5.3 |
| 90th percentile | 7.0 | 5.9 | 6.7 |
| Observed score range | 1.0 to 9.0 | 1.0 to 9.0 | 1.0 to 9.0 |
| Bookings by location: UK / overseas | 5,684 / 4,425 | 3,897 / 2,882 | No combined TMUA total printed |

> The score rows come from Table 4 of the technical report; counts, shares and geography from Tables 1 to 3 of the annual report. The year-on-year column is headed only '% increase from 2024-25', with no statement of whether it is measured on bookings or on delivered tests, so it is reproduced without interpretation. The two documents print the same October figure as 9685 and 9,685; it is written here as 9,685.

## 2. Why the gap between the sittings is not a difficulty gap

1. **Calibrate the items** (Rasch + Winsteps)
TMUA items are calibrated with a Rasch item-response-theory model in Winsteps, producing a difficulty (b) value for each item, with candidate ability written as theta.

2. **Turn raw scores into ability** (raw to theta)
Item difficulties generate a raw-score-to-theta table, and the theta value is then scaled into the reported 1.0 to 9.0 score, given to one decimal place.

3. **Pin two anchors on October** (4.5 and 7.0)
After the October 2025 data was analysed, the median candidate ability of -0.0639 was fixed to a scaled score of 4.5 and the 90th-percentile ability of 1.4634 was fixed to 7.0.

4. **Fit the line, cap the ends** (4.6046 / 1.6369)
A regression line through those two anchors yields the scaling constants 4.6046 and 1.6369, and the transformed scores are capped at 1.0 and 9.0.

5. **January reuses the same constants** (One ruler)
The report states that the same scaling constants were used for both the October 2025 and January 2026 events so that the scaling stayed consistent and scores stayed comparable. Table 3 prints a single row of constants, and its row label is TMUA (Oct 2025).

> The report settles this in one sentence: the same scaling constants were used for both events so that scaled scores stay comparable across them. Its own illustration is that a candidate scoring 6.5, in either January or October, has higher ability than one scoring 4.2 in either event. The ruler did not move, so the distance between means of 4.62 and 3.78 has to come from who sat the test rather than from how hard the paper was. Reading it as a difficulty gap is the single most expensive mistake on this page.

Notice where October's two headline figures land: a median of 4.5 and a 90th percentile of 7.0. Those are not coincidences, they are the two anchor points the scale was pinned to. The October cohort's ability distribution is, by construction, the reference for the whole cycle, and January's 3.6 and 5.9 are that same ruler applied to a different group. Measurement quality is not what separates them either: raw-score Cronbach's alpha runs 0.85 to 0.88 for October and 0.82 to 0.87 for January, above 0.80 on every form.

## 3. Who sits in October, who sits in January

One sentence in the report does a lot of work here, provided you notice what it explains. Candidates may sit the test only once within an admissions cycle, and those applying to the University of Cambridge must sit in the October event, which the report says accounts for the higher volume in that session. That sentence is about volume, not about scores. The score side rests on a different sentence: both events used the same scaling constants. One tells you why the two groups are not the same people, the other tells you the ruler did not change, and the rest follows without anyone having to invent it.

- **How many attempts a cycle allows:** One (The report's wording: candidates are only allowed to sit the test once within each admissions cycle)
- **Cambridge applicants:** Must sit in October (The report attributes the higher October volume to this rule)
- **How big the October window is:** 63.1% (63.1% of the cycle's tests were delivered between 12 and 16 October 2025, counting all three tests)
- **The fee, either sitting:** £75 / £130 (£75 in the UK and Republic of Ireland, £130 anywhere else, quoted once for the cycle)

### The two test windows, and the paper itself

UAT-UK ran three tests across two sittings in 2025-26. Each window consists of six days of testing, with each test delivered across two days, and the annual report notes that October remained the larger of the two windows. The TMUA itself is non-modular and takes the same shape in both events: Paper 1 (Applications of Mathematical Knowledge) and Paper 2 (Mathematical Reasoning), 20 multiple-choice items and 75 minutes each, 150 minutes and 40 items in total, producing one score from 1.0 to 9.0.

### Access arrangements

By sitting, and across all three tests: 801 candidates booked with access arrangements in October, 3.7% of that sitting, of whom 775 attended and sat with the arrangements in place; 652 in January, 5.1%, of whom 617 attended. By type, extra time was 505 in October against 412 in January, rest breaks 74 against 46, both together 202 against 146, and a separate room 44 against 39. For the TMUA on its own, the technical report records 675 candidates with access arrangements, around 4% of the test population, 664 of whom asked for extra time and/or pause-the-clock; the annual report's by-test table, headed as delivered tests, records 676 for TMUA, or 4.2%. The two are counted on different bases, so both are reproduced here and neither is folded into the other.

## 4. The same score, a different position in each group

**TMUA 2025/26 scaled score**: One ruler, two cohorts, 16,092 candidates

- October 2025: 9,685 candidates; the scale is anchored on this event
  - Median 4.5, 90th percentile 7.0
  - Mean 4.62, SD 1.77
- January 2026: 6,407 candidates, using October's scaling constants
  - Median 3.6, 90th percentile 5.9
  - Mean 3.78, SD 1.57
- Whole cycle: 16,092 candidates, the two events pooled
  - Median 4.1, 90th percentile 6.7
  - Mean 4.28, SD 1.74

January's 90th percentile is 5.9 and October's is 7.0. That answers one question only: what it took to reach the top tenth of the group you were sitting with. It does not mean a score is worth more in one month than the other, and the report's own line rules that reading out. Percentiles are descriptive statistics computed within an event; the scaled score is the part that carries across events. The whole-cycle column, with a median of 4.1 and a 90th percentile of 6.7, is the two groups pooled, so do not use it as a benchmark for either sitting on its own.

> Do not treat this table as a way to choose a sitting. There is no easier sitting to be found here: UAT-UK publishes no pass mark, the report contains no score threshold, and because both events share one set of scaling constants, switching months is not a route to a higher scaled score. What matters is what your target universities require and how far along your preparation is. For dates and booking see [TMUA registration and dates](/exams/tmua/guides/registration-and-dates), and for how the score is built see [how TMUA is scored](/exams/tmua/guides/scoring).

## 5. What these numbers cannot settle, and how to decide

### A percentile is not a threshold

UAT-UK publishes no pass mark, and neither report contains any university threshold. A 90th percentile of 7.0 says that a tenth of that event's candidates reached or passed that score; it does not say that any institution treats it as a line. The reports also carry no admissions outcomes, so this page derives no acceptance rate and turns no percentile into a threshold. For the full distribution and the group breakdowns see [TMUA score statistics](/exams/tmua/guides/score-statistics).

### These scores do not compare across cycles

This cycle's scaling constants were fixed on the October 2025 data and then used for both events. The scale therefore belongs to the 2026 admissions cycle, and placing a score from an earlier cycle beside it has no basis in the report. If you want to locate yourself, the reference is this cycle's two distributions, not a number passed down from an earlier year.

### Timing data is not split by sitting

The test-time table, Table 18, prints a single row covering the 15,481 candidates included in the analysis: a mean of 02:27:01 and a median of 02:29:37 against a standard limit of 02:30:00, with 610 candidates excluded for exceeding 02:30:04 and assumed to have a time-based access arrangement. A total of 91% of candidates used between 02:15:00 and 02:30:04. Unreached items are reported by paper rather than by event: 322 candidates on Paper 1 and 135 on Paper 2, a very small share of the 16,091 who took the test. If you wanted to compare pacing pressure between the two sittings, the printed figures do not support it.

### Chart-only values are left out

Several exhibits in these reports exist only as images: the binned score distribution, the share of candidates by test time, and the annual report's by-country bar chart split between October and January. Their values are not printed as text, and a number estimated by eye off a chart only looks precise. Every figure on this page can be matched word for word to a printed table or sentence; anything that lives only inside an image is left out.

So the order of decisions runs like this. First check what each university and course you are applying to actually requires of the TMUA sitting; the report is explicit that Cambridge applicants must sit in October. Then weigh how ready your preparation is. Only after that does the score distribution enter the picture at all. You may sit once per cycle, there is no second attempt inside the cycle, and the fee is the same either way, so this is a one-shot decision and it should not rest on the distance between two means.

> Every figure here comes from a printed table or sentence in UAT-UK's 2025/26 TMUA technical report and its 2025-26 annual report. Nothing is read off a chart image and nothing is summed or averaged across tables. The reports contain no admissions outcomes, so this page computes no acceptance rate. The annual report separately records that tests were taken in 129 different countries in 2025-26.

## Questions
**October averaged 4.62 and January 3.78. Was the January paper harder?**

No. A raw score reflects both candidate ability and form difficulty, while the scaled score removes form difficulty. UAT-UK applied the same scaling constants, 4.6046 and 1.6369, to both events, and the report says this was done so that scaled scores stay comparable within the cycle. The distance between 4.62 and 3.78 therefore reflects a difference in who sat, not in how hard the paper was.
**So should I sit in October or in January?**

Start with what your universities require, which often settles it: the report states that candidates applying to the University of Cambridge must sit in October, and you may sit only once per cycle. If both events are accepted for your courses, the remaining question is how ready your preparation is, not how the two means compare. The fee is identical (£75 in the UK and Republic of Ireland, £130 elsewhere), and the format, item count and timing are the same in both.
**Is 5.9 in January equivalent to 7.0 in October?**

No. 5.9 is the 90th percentile of the January group and 7.0 is the 90th percentile of the October group, so both describe the entrance to the top tenth of the people you sat with. The scaled score itself is one ruler across events: the report's example is that a candidate scoring 6.5, in either month, has higher ability than one scoring 4.2 in either event. The percentile moves with the group; the score does not.
**Were the two sittings identical in format and measurement quality?**

The format is identical. The TMUA is non-modular: Paper 1 (Applications of Mathematical Knowledge) and Paper 2 (Mathematical Reasoning), 20 multiple-choice items and 75 minutes each, 150 minutes and 40 items in total, producing one score from 1.0 to 9.0. Measurement quality is close too: raw-score Cronbach's alpha runs 0.85 to 0.88 in October and 0.82 to 0.87 in January, with scaled-score reliability 0.85 to 0.87 and 0.82 to 0.87 respectively, every form above 0.80.
**Candidate numbers are rising. Does that change my score?**

Not your scaled score. October delivered 9,685 tests and January 6,407; the annual report's year-on-year column reads +8% for October and +32% for January, so January grew considerably faster. But the scale was fixed on the October data and both events share the same constants, so a score is not pushed down by a larger cohort sitting alongside you. Growth changes how the competitive picture is described, not how your score is computed.

## Official sources
- [UAT-UK TMUA Technical Report 2025-26, Table 4: scaled score summary statistics by event](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161815/UAT-UK-TMUA-Technical-Report-2025-26.pdf)
- [Same report, Tables 2 and 3: anchor ability estimates and the scaling constants shared by both events](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161815/UAT-UK-TMUA-Technical-Report-2025-26.pdf)
- [Same report, Tables 16 to 18: reliability and SEM by event, and whole-cohort test time](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161815/UAT-UK-TMUA-Technical-Report-2025-26.pdf)
- [UAT-UK Annual Report 2025-26 Candidate Data, Tables 1 to 3: candidates, year-on-year change and geography by sitting](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161742/Annual-Report-2025-26-Candidate-Data.pdf)
- [Same annual report, Tables 6 to 8: access arrangements by sitting and by test](https://uat-wp.s3.eu-west-2.amazonaws.com/wp-content/uploads/2026/08/07161742/Annual-Report-2025-26-Candidate-Data.pdf)
- [UAT-UK Annual Reports index, the public entry point for the 2025-26 reports](https://esat-tmua.ac.uk/about-uat-uk/annual-reports/)
