UFC Vegas 123 AI crowd: Claude models back Nolan, GPT models back Moicano

The AI crowd leans towards Tom Nolan over Renato Moicano in Saturday's main event, but only just. Nolan got 59 of 100 picks, and the average chance came out at 51%.

The crowd is 100 AI forecasters, played by three Claude models from Anthropic and two GPT models from OpenAI, all reading the same fact sheet with no odds and no web search.

At UFC 333: Volkanovski vs. Evloev, the crowd got 7 of 14 fights right. Our model got 9 of 14, and the betting favourites won 9 of 11. The crowd's biggest miss was Asaf Chopurov, backed at 70%, who lost a three-round decision to Colby Thicknesse.

The main event splits by company

The 51% hides a near-total split. Claude forecasters picked Nolan 55 times out of 60, at 57%. GPT forecasters went the other way, with 36 of 40 on Moicano at 58%.

All 20 Claude Opus 5.5 forecasters took Nolan. GPT-6.1 Sol was almost as one-sided in the other direction, with 19 of 20 for Moicano.

The Claude side built its case on age and size. "At 26 and 6'3", Nolan's volume (4.72 landed a minute) and submission threat should outlast a 37-year-old Moicano over five rounds," wrote a Claude Opus 5.5 forecaster weighing the stand-up.

The GPT side saw a route on the mat. "Nolan's 58% takedown defence offers openings for Moicano's proven top control and submissions, though securing takedowns remains the key uncertainty," said a GPT Astra forecaster weighing the grappling.

The outside numbers are just as torn. The betting market gives Moicano 52%, and our model gives Nolan 51%.

Where the crowd and the odds part ways

The biggest gap is Jean-Paul Lebosnoyani against Farman Hasanov. The betting market and our model both have Lebosnoyani at 72%, but the crowd gives him 54%. He was still picked by 78 of 100 forecasters, and the ten weighing the grappling were the only angle group to side with Hasanov, 7 of 10.

The reasons given for Lebosnoyani kept coming back to Hasanov's cardio. "Hasanov visibly gassed in round two of his debut; Lebosnoyani's pace and submission threats punish fading wrestlers," wrote a Claude Fable 5.1 forecaster weighing age, time away and the scale.

Julian Erosa against JeongYeong Lee came out dead even, so the crowd has no pick. Erosa got 53 picks to Lee's 47. The odds give Lee 60% and our model 61%. Claude forecasters leaned Lee, 36 of 60, while GPT forecasters took Erosa, 29 of 40.

The durability group went 9 of 10 for Lee. "Erosa's chin has aged (6.23 strikes absorbed a minute, recent first-round KO) and he is retiring; Lee carries ring-rust and weight-cut risk," wrote a Claude Sonnet 5.5 forecaster weighing durability and cardio.

Ode Osbourne against Cody Durden runs the other way. The betting market has it even and our model gives Durden 52%, but 93 of 100 forecasters picked Durden, at 57%.

Seven unanimous fights

Every forecaster agreed on seven fights: Jacobe Smith, Luana Santos, Francis Marshall, Joe Kropschot, Piera Rodriguez, Lucia Szabova and Zach Reese. Even so, the crowd's top number is Szabova's 72%. The odds have her at 83% and our model at 89%.

Our read is that the crowd stays cautious even when it agrees. It never went above 72% on this card, while the betting market went as high as 83%.

The GPT models were the bolder group on two of these. They gave Reese 76% against 65% from the Claude models, and Rodriguez 70% against 60%.

Our model and the betting market as of Oct 27, 2:34 PM ET.

How the AI crowd works

The crowd is 100 AI forecasters, played by Claude Opus 5.5, Claude Sonnet 5.5 and Claude Fable 5.1 from Anthropic, and GPT Astra and GPT-6.1 Sol from OpenAI. Most were told to weigh the fights from one of nine angles: the stand-up, the grappling, the numbers, durability and cardio, age and time away, recent form, the matchup, the sceptic's case and the judging. The rest got no angle at all, only the fact sheet and the question.

All of them read the same fact sheet: each fighter's record, size, career stats and recent UFC fights from our database, plus this week's news, gathered once for each fight. None of them saw the betting odds, our model or each other's picks, and none could search the web. Half of them saw each fight with the fighters in the opposite order, so the order they're listed in can't tilt the result.

Each forecaster picked a winner in every fight and gave a chance. The crowd's number is the average of those chances. It's an experiment in how AI reads fights, not betting advice.