AI crowd · UFC 334: Gane vs. Hokit
AI crowd picks Gane at UFC 334: Gane vs. Hokit and breaks with the odds on Thompson
All 10 AI forecasters picked Ciryl Gane to beat Josh Hokit, at 63%. They were just as united in picking Charles Radtke over Stephen Thompson, the betting market's favourite.
The setup
10 AI forecasters, played by five AI models, each playing 2. Each one reads the same fact sheet on every fighter (record, size, career stats, UFC fights and this week's news) and gives a winner and a chance for every fight. None of them sees the betting odds, our model or each other's picks, and none can search the web. The crowd's number is the average of their chances.
The models
- Claude Sonnet 5.5Anthropic · 2 forecasters
- Claude Opus 5.5Anthropic · 2 forecasters
- Claude Fable 5.2Anthropic · 2 forecasters
- GPT AstraOpenAI · 2 forecasters
- GPT-6.1 SolOpenAI · 2 forecasters
All 10 AI forecasters picked Ciryl Gane to beat Josh Hokit in Saturday's heavyweight title fight, and their average chance for him was 63%. That matches the betting market exactly and sits above our model's 56%.
The crowd is 10 forecasters played by five AI models, three from Anthropic and two from OpenAI, all reading the same fact sheet without seeing the odds.
Last week went well for the crowd. On UFC Fight Night: Cortes Acosta vs. Walker, it got 11 of 14 fights right, against 10 for our model and 9 for the betting favourites. Its biggest miss was Mantas Kondratavicius at 67%, whom Andre Petroski stopped in round 2. It also had Waldo Cortes Acosta at 64% in the main event, and Johnny Walker stopped him in round 3.
All 10 against the betting favourite
Stephen Thompson is the betting market's favourite against Charles Radtke at 58%, and our model leans the same way at 56%. The crowd went the other way. All 10 forecasters picked Radtke, at 65%, with the GPT models at 68% and the Claude models at 63%.
Their reasons were mostly about age, time away and wrestling. A Claude Fable 5.2 forecaster wrote: "Thompson is 43, off a 490-day layoff with knee, neck and shin injuries, and recent opponents have out-wrestled him."
Where the models split
Evan Elder vs Samuel Sanches is the fight the crowd found closest. Seven of 10 forecasters picked Elder, at 54%, while the betting market and our model both lean slightly to Sanches, at 52% and 53%.
The split ran along company lines. All 4 GPT forecasters took Elder, at 61%, while the Claude forecasters split 3 and 3. Both Claude Opus 5.5 forecasters took Sanches, and one of them wrote: "Sanches is a 23-year-old knockout puncher with an 11-1 record; Elder has been out 567 days and his takedown defence is 40%."
Uros Medic vs Kevin Holland was the other fight without a unanimous pick. Eight of 10 took Medic, at 55%. The two who didn't were both GPT Astra forecasters, which gave Holland 54%. One of them argued: "Holland's ten-inch reach advantage and submission threats offer answers to Medic's explosive striking; Medic has already lost twice by submission."
Right side, less sure
The crowd's highest number on the card is 74%, for Bilal Hasan over Luis Gurule. That is well short of the betting market's 87% and our model's 92%. The same gap shows on Juan Diaz, at 64% against 75% in the odds and 82% from our model. Melquizael Costa gets 65% from the crowd, 76% from the odds and 80% from our model.
On Diaz, the forecasters saw the edges but stayed cautious. A Claude Fable 5.2 forecaster wrote: "Diaz had a full camp, a size edge and five takedowns in his debut; Alatengheili steps in on eight days' notice."
Donte Johnson is the exception. The crowd gives him 70% against Cesar Almeida, above the betting market's 60%.
| Fight | AI crowd | Our model | Betting market |
|---|---|---|---|
| Gane vs Hokit | Gane 63% | 🔒 | Gane 63% |
| Harrison vs Nunes | Harrison 58% | 🔒 | Harrison 62% |
| Borralho vs Belgaroui | Borralho 61% | 🔒 | Borralho 58% |
| Medic vs Holland | Medic 55% | 🔒 | Medic 59% |
| Hasan vs Gurule | Hasan 74% | 🔒 | Hasan 87% |
| Thompson vs Radtke | Radtke 65% | Thompson 56% | Thompson 58% |
| Dober vs Duncan | Duncan 63% | Duncan 68% | Duncan 60% |
| Miller vs McKinney | McKinney 60% | McKinney 70% | McKinney 66% |
| Johnson vs Almeida | Johnson 70% | Johnson 68% | Johnson 60% |
| Gordon vs Costa | Costa 65% | Costa 80% | Costa 76% |
| Chiasson vs Mesquita | Mesquita 68% | Mesquita 74% | Mesquita 69% |
| Alatengheili vs Diaz | Diaz 64% | Diaz 82% | Diaz 75% |
| Elder vs Sanches | Elder 54% | Sanches 53% | Sanches 52% |
Our model and the betting market as of Nov 12, 7:50 AM ET.
Every forecaster, fight by fight
Each dot is one AI forecaster, placed by the chance it gave its pick. Left of the middle line it picked the red-corner fighter, right of it the blue corner. Dots pile up in 5-point steps, Claude models above the line and GPT models below it.
Model by model
Model by model
Each model's own call, from its forecasters given no angle, next to the whole crowd. The number is the chance it gave its pick; red means it picked the red-corner fighter, blue the blue corner, and the deeper the colour, the surer it was.
| Fight | AI crowd | Opus 5.5 | Sonnet 5.5 | Fable 5.2 | GPT Astra | GPT-6.1 Sol |
|---|---|---|---|---|---|---|
| Gane v Hokit | 63 for Gane | 62 for Gane | 62 for Gane | 62 for Gane | 65 for Gane | 64 for Gane |
| Harrison v Nunes | 58 for Harrison | 56 for Harrison | 60 for Harrison | 60 for Harrison | 57 for Harrison | 56 for Harrison |
| Borralho v Belgaroui | 61 for Borralho | 60 for Borralho | 57 for Borralho | 62 for Borralho | 62 for Borralho | 62 for Borralho |
| Medic v Holland | 55 for Medic | 57 for Medic | 61 for Medic | 56 for Medic | 54 for Holland | 56 for Medic |
| Hasan v Gurule | 74 for Hasan | 70 for Hasan | 72 for Hasan | 80 for Hasan | 73 for Hasan | 76 for Hasan |
| Thompson v Radtke | 65 for Radtke | 60 for Radtke | 66 for Radtke | 64 for Radtke | 66 for Radtke | 68 for Radtke |
| Dober v Duncan | 63 for Duncan | 61 for Duncan | 60 for Duncan | 62 for Duncan | 65 for Duncan | 66 for Duncan |
| Miller v McKinney | 60 for McKinney | 61 for McKinney | 56 for McKinney | 66 for McKinney | 57 for McKinney | 60 for McKinney |
| Johnson v Almeida | 70 for Johnson | 69 for Johnson | 72 for Johnson | 72 for Johnson | 68 for Johnson | 68 for Johnson |
| Gordon v Costa | 65 for Costa | 61 for Costa | 65 for Costa | 70 for Costa | 64 for Costa | 64 for Costa |
| Chiasson v Mesquita | 68 for Mesquita | 68 for Mesquita | 66 for Mesquita | 68 for Mesquita | 68 for Mesquita | 70 for Mesquita |
| Alatengheili v Diaz | 64 for Diaz | 66 for Diaz | 62 for Diaz | 70 for Diaz | 60 for Diaz | 60 for Diaz |
| Elder v Sanches | 54 for Elder | 56 for Sanches | 55 for Elder | 50 for neither | 60 for Elder | 62 for Elder |
How the AI crowd works
The crowd is 10 AI forecasters, played by Claude Opus 5.5, Claude Sonnet 5.5 and Claude Fable 5.2 from Anthropic, and GPT Astra and GPT-6.1 Sol from OpenAI. None of them was told how to weigh the fights: each got only the fact sheet and the question.
All of them read the same fact sheet: each fighter's record, size, career stats and UFC fights from our database, plus this week's news (the reports, and what the fighters and their coaches said), 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.