UFC Shanghai predictions: our model fades the favorites
Our AI model disagrees with the betting market on Yan Xiaonan vs Denise Gomes and fades the favorite in eight of the thirteen UFC Shanghai fights.
Our prediction model has now scored all thirteen bouts at UFC Shanghai. It disagrees with the betting market about who wins two of them, and it is quietly less impressed than the market with almost every favorite on the card.
That second part is the more interesting finding, so let me show the work before I make anything of it.
What the two numbers are
The model gives every upcoming UFC bout a win probability for each corner, rebuilt from our database and stamped against the two fighters actually booked. The market number is the moneyline, converted to an implied probability and then stripped of the sportsbook's margin, so the two sides of a fight sum to 100 percent and the comparison is apples to apples. Odds below were pulled Thursday morning.
Here is every fight on the card, model probability against devigged market probability, from the perspective of the fighter listed first.
| Bout | Model | Market | Gap |
|---|---|---|---|
| Umar Nurmagomedov vs Song Yadong | 74.2% | 83.4% | -9.2 |
| Yan Xiaonan vs Denise Gomes | 37.2% | 58.0% | -20.8 |
| Aoriqileng vs Kai Asakura | 25.9% | 20.9% | +5.0 |
| Sumudaerji vs Alex Perez | 60.2% | 66.4% | -6.2 |
| Ce Liu vs Levi Rodrigues Jr. | 48.9% | 63.3% | -14.4 |
| Bilal Hasan vs Nilson Rojas | 62.4% | 85.8% | -23.4 |
| Namsrai Batbayar vs André Lima | 27.7% | 30.0% | -2.3 |
| Rei Tsuruya vs Kevin Borjas | 78.2% | 84.7% | -6.5 |
| Jack Jenkins vs Sean Woodson | 41.8% | 42.9% | -1.1 |
| Xiao Long vs Francesco Nuzzi | 64.8% | 61.8% | +3.0 |
| Lawrence Lui vs Hector Santiago | 72.4% | 70.9% | +1.5 |
| Xiong Jing Nan vs Julia Polastri | 41.5% | 31.6% | +9.9 |
| Ding Meng vs Cam Nelson | 75.5% | 57.1% | +18.4 |
The two outright disagreements
Yan Xiaonan (19-5) is a modest betting favorite over Denise Gomes (12-3) at -150. She is ranked fourth at women's strawweight and twelfth in the women's pound-for-pound list; Gomes is ranked fourteenth. Every ordering signal on the page says Yan.
The model has Gomes winning 62.8 percent of the time.
That is a twenty-point break from the market on a fight between ranked strawweights, and it is not a rounding artifact of a thin sample. Yan has been in the UFC since 2018 and Gomes since 2022, so both have real records for the model to chew on. Whatever it is reading, it is reading it deliberately.
The other flip is quieter. Ce Liu is a -190 favorite against Levi Rodrigues Jr., and the model calls it 51.1 percent for Rodrigues. Ce Liu is 3-1 as a professional and Rodrigues is 5-0, which is exactly the sort of matchup where a market prices the name it recognizes and a model prices the record in front of it.
Photo: Новости ННТ / CC BY 3.0
The bigger pattern is the fading
Look down the gap column again and count the negatives. On eight of the thirteen fights the model sits below the market on the fighter the market favors, and every single fade larger than five points lands on a favorite priced at -150 or shorter.
The widest gap on the card is not one of the flips. It is Bilal Hasan against Nilson Rojas, where the market has Hasan at 85.8 percent and the model has him at 62.4. Both men are undefeated, 9-0 and 8-0. A -850 line on an undefeated fighter facing another undefeated fighter is the market saying it knows something about the quality of those records; the model, which has almost nothing to go on for either man, is saying it does not.
That is worth naming plainly, because it cuts against the piece. A model that has never seen a fighter compete at this level will regress him toward the middle, and "regress toward the middle" looks identical to "fade the favorite" when you print it in a table. Two of the four widest gaps on the card, Ce Liu and the Ding Meng fight at the bottom of the prelims, involve a man with four or fewer professional bouts in our database.
The Yan fight is the one that survives that objection. Both women have been fighting at this level for years. There is no thin-sample excuse available, which is what makes it the number worth watching on Saturday.
The one the model likes more
Xiong Jing Nan is the ONE strawweight champion. She fought twelve times in that promotion between 2017 and 2025, and nine of those twelve were world title fights. In May she made her UFC debut and lost a unanimous decision to Angela Hill.
The market has taken that one result seriously: she is a +205 underdog against Julia Polastri on the prelims. The model gives her 41.5 percent, ten points more generous than the line.
Ten points is not a scream. It is the model declining to reprice nine title fights on the strength of one bad night, which is roughly what a rating is for.
Then Saturday happens
We have been scoring these against results for a while now, most recently on the Sacramento card, and the honest summary is that neither side of this comparison has earned the right to be smug. The market is very good. The model is occasionally better, in a specific way: it is willing to be wrong loudly on a fight where the crowd is confident.
Yan Xiaonan is fighting in China, ranked ten places above her opponent, and our own model thinks she loses six times out of ten. One of us is going to look silly at 6:00 a.m. Eastern.