MMA Career Rankings

Rankings

The project publishes All-time and Prime career rankings for men and women from the UFC, PRIDE, WEC, Strikeforce, Affliction, Bellator and RIZIN. The rating model uses recorded professional fights across all promotions and the regional circuit. Current and weight-class boards cover UFC fighters.

The interactive site filters boards and analysis by roster, weight class, board size and minimum rated bouts.

Rating model

Design choice	Rationale
Whole-History Rating (Coulom, 2008)	A one-pass rating like Elo never goes back. If an opponent you beat later becomes a champion, that win should count for more, and here it does. Fitting all careers together also keeps fighters from different decades on one scale.
A neutral starting bout	Without it, a fighter at 2-0 has no upper limit and can top the board. With it, an unbeaten record has to keep growing against real opposition to climb.
Graded results	A knockout says more about who was better than a split decision does.
A contender line for each year	The rating scale drifts upward over time. Measuring against that year's contenders compares a 2007 win and a 2023 win fairly.
Win probability raised to a power	One win over an elite contender is worth about five wins over borderline ones, so a long run of borderline wins cannot outscore a few elite ones. The third power was chosen after tests of powers from 1 to 6.
Promotion weights, pulled toward average	Keeping promotion out of the rating lets results decide how strong each pool was. The weight then accounts for depth of competition, and pulling thin evidence toward the average stops a handful of labelled fights from deciding a whole career.
Square roots on a common scale	It puts the three parts on one scale and rewards a career that is strong in all three more than one huge number in a single part.
Best ten-year window	It measures a peak without punishing a fighter for a long career before or after it.

The model has not yet been shown to predict fights better than a simpler rating; that test is recorded as open in the release check. The rankings are a measured judgement of careers, not a betting model.

Site features

Feature	What you can do
Rankings	All-time, Prime and Current, exactly as published below. Pick a weight class for that class's own board: the All-time score earned in UFC bouts in that class.
Weight-class boards	UFC only. A champion in two classes is ranked in both, with the share of their All-time score earned in each.
Score anatomy	See each score split into championships, career skill and contender résumé. Move two sliders to change the exchange rate and watch the board re-rank.
Rating method	Each method behind the rankings, what it does and why it was chosen, in ordinary words.
Career arcs and tale of the tape	Overlay up to six fighters' ratings fight by fight and compare their records, ranks and class placements side by side.
Where a career score comes from	A year-by-year receipt that sums to the career score.
Risers, fallers and streaks	The biggest rating moves of any year, and the longest UFC win streaks placed by length and by the strength of who they beat.
Biggest upsets	The UFC wins the betting market rated least likely, and how often favourites win.

Each chart includes a finding calculated from the data under the current controls. The page follows the reader's light or dark setting.

On a phone, open Explore the rankings to change the filters. Wide tables scroll inside their panels. Weight-class buttons work with a keyboard, and the page follows the reader's reduced-motion setting.

To check the interface locally, run npm ci, npx playwright install chromium, then npm test. The browser checks cover search positions, roster filters, fighter comparisons, analysis charts, phone layouts and dark mode. They also run in CI.

Weight-class board	Score anatomy and exchange rates
Career arcs and tale of the tape	Weight classes over time

Model rules

Weight classes are separate pools. Career division is where a fighter competed most; current division changes only after a UFC title win. Each year is measured against its division's top ten percent, roughly the fifth-best fighter; classes with fewer than 30 fighter-years use the sport-wide line. Men and women remain separate.

The fight rating ignores promotion. Exposure is applied later: UFC is 1.00, PRIDE 0.95, Affliction 0.90, WEC and Strikeforce 0.88, Bellator 0.65, and PFL, RIZIN, ONE and DREAM 0.60. Thin evidence shrinks toward the average.

A knockout or submission scores 1.00, a unanimous decision 0.95, a split or majority decision 0.90, a disqualification 0.85, and a draw 0.50 for each fighter; no-contests are excluded. Careers are fitted together, so later results can revalue earlier fights. Ratings sit around 1,500, and a neutral virtual bout checks tiny unbeaten records. Duplicate fights are removed and whole careers are required.

What the score measures

All-time = 1,000 × (30% √title + 17.5% √skill + 52.5% √résumé)

Each part is divided by the average of its own top 100, then square-rooted, before weighting. A part of 1.0 still means "as much as a typical top-100 career", but doubling a part adds only about 41%, so a career is paid for breadth across the three parts more than for an extreme in one.

Part	Weight	What it measures	Exact rule
Title	30%	Title fights won, valued by who was beaten	Each title win is worth q⁴, where q is the chance the opponent beats a contender-level fighter, using the opponent's rating going in against that division's contender level that year. A major-promotion title win is worth at least 0.05.
Résumé	52.5%	Wins over contenders, valued by how strong they were	Counts a win over an opponent rated at least 1,750 going in, with eight or more UFC bouts. Each is worth q³ against 1,750: 0.13 just over the line, 0.26 at 100 points above, 0.61 at 300 above. At most 1.0 per calendar year.
Skill	17.5%	How far the fighter's own rating stood above contender level	For every year with a fight: the fighter's average rating that year minus the division's contender level that year (about its fifth-best fighter). Years below the line add almost nothing; years above add the excess. Summed over the career.

The rating behind all three is one model of every recorded fight, so beating a fighter counts more the better that fighter proved to be. Skill and résumé are then multiplied by the promotion exposure above, between 0.20 and 1.00. An opponent returning from a long absence is marked down 90 points per excess layoff, up to four. Because q³ rises steeply, one win over an opponent 300 points above the line is worth about five wins over borderline contenders.

All-time and Prime read every rating against its own year's contender line. The model's scale drifts upward over time: on this fit the 60th-best man stood at 1,834 in 2007 and 1,919 in 2023, so without this era adjustment a 2023 win would be priced on a scale about 85 points higher than a 2007 win. Current, whose fighters are all contemporaries, reads the unadjusted ratings.

The score is a points total, not a ratio. Jon Jones's 1,814 against Georges St-Pierre's 1,564 does not mean Jones was 16% better. It means his three parts add up to more points: title 617 against 482, résumé 812 against 777, skill 385 against 304. Most of the résumé gap is two elite wins: Daniel Cormier at 334 points above the line and Ciryl Gane at 261, where St-Pierre's best was Sean Sherk at 189. For who was better at their best, read Prime: 2,193 against 2,110, an 83-point edge, which the rating turns into about a 62% chance of winning.

Prime takes the ten-year stretch with the most wins over contenders, using the same contender rule as the résumé, and needs four. It ranks by those wins times the stretch's average rating above the lowest qualifying average. Current uses the All-time formula over the last 18 months: title wins and contender wins in that window, plus the latest rating above 1,750, capped at the best opponent the fighter has ever beaten so an unbeaten run against weak opposition cannot keep climbing. It needs 13 rated appearances, eight UFC bouts and a UFC bout within 18 months. Integrity deductions never alter a published board.

Each weight class has its own board, and it is UFC only: the All-time score counted only from the title wins, contender wins and career years earned in UFC bouts in that class. All-time and Prime count every promotion; the class boards and Current count the UFC. What a fighter earned outside the UFC stays in their All-time score but in no class, so their class scores add up to their All-time score less that remainder. A champion in two classes is ranked in both.

Published rankings

The men's headline boards are published below so the rankings are visible on the repository front page. The full Published MMA Rankings adds the three women's boards and the weight-class boards.

Release fact	Value
Dataset	2026-10-24
Data through	2026-10-24
Included records	UFC, early UFC and major-promotion careers
All-time basis	All-time career score
Rated fights	81,643
Rated fighters	34,150
Available fight records	83,053
Whole-career coverage	1,827 of 1,840 eligible fighters (99.3%)
Prime contender threshold	1,750 — reached by 25.3% of established fighters
Prime qualifiers	102 men; 7 women

All-time — men, top 100

Prime and Prime rank come from the Prime board below, and Elite wins is the evidence behind that rank. A blank rank means the fighter did not qualify, not that they placed last.

#	Fighter	Division	Score	Prime	Prime rank	Elite wins
1	Jon Jones	Light Heavyweight	1813.7	2193	1	11
2	Georges St-Pierre	Welterweight	1563.5	2110	2	12
3	Islam Makhachev	Lightweight	1559.8	2157	4	8
4	Alexander Volkanovski	Featherweight	1406.2	2032	6	10
5	Anderson Silva	Middleweight	1384.4	1955	3	14
6	Jose Aldo	Featherweight	1364.0	1915	12	9
7	Daniel Cormier	Light Heavyweight	1348.0	2055	7	8
8	Demetrious Johnson	Flyweight	1328.9	1951	41	5
9	Stipe Miocic	Heavyweight	1214.4	1951	24	6
10	Dominick Cruz	Bantamweight	1208.4	1933	45

Prime, elite-tested — men, top 50

The table prints both inputs instead of the internal ordering index.

#	Fighter	Division	Prime	Elite wins
1	Jon Jones	Light Heavyweight	2193	11
2	Georges St-Pierre	Welterweight	2110	12
3	Anderson Silva	Middleweight	1955	14
4	Islam Makhachev	Lightweight	2157	8
5	Fedor Emelianenko	Heavyweight	2085	9
6	Alexander Volkanovski	Featherweight	2032	10
7	Daniel Cormier	Light Heavyweight	2055	8
8	Antonio Rodrigo Nogueira	Heavyweight	2041	7
9	Frankie Edgar	Lightweight	1911	10
10	Max Holloway	Featherweight	1924	9
11	Lyoto Machida	Light Heavyweight	2011	7
12	Jose Aldo	Featherweight	1915	9
13	Sean Sherk	Lightweight	1984	7
14	BJ Penn	Lightweight	1888	9
15

Current — men, top 30

#	Fighter	Division	Score	Title wins	Contender wins	UFC bouts	Last UFC bout
1	Justin Gaethje	Lightweight	1902	2	2	16	2026-06-14
2	Islam Makhachev	Welterweight	1847	2	2	19	2026-08-15
3	Joshua Van	Flyweight	1793	3	4	12	2026-09-19
4	Alexander Volkanovski	Featherweight	1735	2	2	19	2026-10-24
5	Petr Yan	Bantamweight	1721	2	2	17	2026-10-24
6	Sean Strickland	Middleweight	1610	1	2	25	2026-05-09
7	Merab Dvalishvili	Bantamweight	1455	2	2	18	2026-10-24
8	Khamzat Chimaev	Middleweight	1428	1	1	10	2026-05-09
9	Ilia Topuria	Lightweight	1359

Limitations

All-time is retrospective, elite wins reward opportunity, and the age curve misses decline in fighters who never returned. Promotion labels are missing from 54% of rated fights; 11% of filled weight classes are estimated wrong. Whole-career coverage is complete for every eligible fighter, and none has an evidenced source-ID conflict. See the decision register.

Repository files

Path	Use it for
RANKINGS.md	Full rankings
ratings/ / loaders/	Model / data assembly
ratings/fights.py, divisions.py, history.py	Shared fight preparation, division names and pre-fight lookups
ratings/evaluation.py, bootstrap.py, boards.py	Probability metrics, interval setup and published board definitions
rust/whr_rs/	Whole-History Rating fitting and conditional variances in Rust
rust/rating_rs/	Glicko-2 updates and inactivity transitions in Rust
data/snapshots/<date>/	The release: inputs, fitted ratings and boards, all versioned
docs/	Method and decisions
docs/FORECAST.md	The pre-fight forecast benchmark against betting odds, and the selection leak it avoids
docs/RELEASE_CHECK.md	Source freshness, fresh-checkout reproduction and the limits of current predictive evidence
site/	Interactive analysis site, published at https://symonyeal.github.io/MMA-Career-Rankings/; open site/index.html locally
tests/	Verification

Use the system Python:

python -m pip install -r requirements.txt
python -m pip install maturin ./rust/whr_rs ./rust/rating_rs
python -m pytest -q
python check_release.py
python build_boards.py data/snapshots/2026-10-24 --scope majors,pre_unified --write-readme

The second line requires a Rust toolchain (rustup). Both compiled extensions are required: there is no Python solver or fallback. Tests check captured numerical results and model behavior.

Refresh the narrative figures after rebuilding a release: python analysis/readme_figures.py --write-readme. The tests check these figures against the data.

The site draws its analyses from one snapshot. Its data file, site/data.js, is generated; rebuild it after the boards change:

python analysis/build_site.py data/snapshots/2026-10-24

Use --output <directory> to build an isolated copy for review. The retired notebook and its builders are preserved in _archive/20261029-notebook-retired/.

index.html loads each asset as name?v=<content hash>, so a browser never pairs the page with a cached older asset. After editing app.js or style.css alone, re-stamp with python analysis/build_site.py --stamp; a test fails while the stamp is stale.