Engine online · calibrated for UFC

Powered by theAPEXprediction engine

Most predictions ask for your trust and show you nothing. This one shows its work. Every fight gets a win probability, a likely method and a finish timing, read from the fighters' own numbers — and opens up so you can check the reasoning yourself.

Run the engine Browse events

Most confident 25%

71.1%

Bouts back-tested

3,307

Confidence cap

85%

Next event

Dana White's Contender Series: Season 10, Week 5

Tuesday, September 8, 2026 · 7:00 PM EDT

Main event

Light Heavyweight · 3 rounds

Quentin
Pasley

3-0-0 pro

Arlind
Berisha

5-0-0 pro

Strikes / min——

Absorbed / minlower wins——

Accuracy——

Takedowns / 15 min——

Takedown defense——

Finish rate——

Pasley Berisha

Inside the app

01 / Schedule

Events

Full fight cards, main to early prelims, with live schedules and results.

Open Events

How it works

Most UFC picks are sold as locks. Edge Engine does the opposite: it gives you the honest probability of each outcome — and shows exactly how it got there — so you can read the fight for yourself.

Its own read, not the market's

The number comes from the fighters' own record, output and durability — not from the betting line. The market is shown beside it so you can see where the two disagree.

Calibrated, not cocky

Probabilities stay honest — nothing here is ever a lock. A 60% pick is a 60% pick, and the record shows how those have actually landed.

It shows its work

Win, method, and — when both fighters have the UFC record to earn it — a specific round, each with the reasoning behind it. On the rest it says so rather than guessing. That engine is APEX.

And it's kept honest: every card is locked before the fights and graded in public, wins and losses alike. See the track record →

Does it actually work?

Tested against history

61%

winners called over 3,307 past bouts, each predicted using only what was known beforehand

The full record →

Winners calledvs coin flip 50.0% · higher is better

61.0%50.0%

Brier scorevs coin flip 0.250 · lower is better

0.23310.250

Method calledvs a 1-in-3 guess 33.3%

50.6%33.3%

Round calledvs 33.3% · on the 30.4% it commits to

51.3%33.3%

Each bar has its own scale and doesn't start at zero; the pale mark is the number to beat. The Brier bar is short on purpose — that's the one where a lower score is better. Rounds count only the 1006 bouts the engine named a round on, 30.4% of them. It stays quiet on the rest. UFC fights only. The engine was trained on older fights, then tested on newer ones it had never seen.