Brazil 2-1 Japan. AI World Cup Predictions: The 'Boldest' AI, Gemini, Was the Most Accurate | AI×SPORTS LAB Part 4: Results Review
A series where ChatGPT, Claude, Gemini, and humans predict the same match under the same conditions and scoring criteria, then review the results with data after the match. The results for Part 4, the Brazil match, are in.
The result was Brazil 2-1 Japan.
Japan was eliminated in the first round of the knockout stage.
The most accurate predictor this time was Gemini (Brier 0.23).
Claude maintains the overall lead. This is the exact opposite of the previous match (a draw). The more strongly an AI backed Brazil, the more accurate it was.
The conclusion first
After scoring the fourth installment of AI×SPORTS LAB, the most accurate this time was Gemini. This is because it backed Brazil the most strongly among the three (61.5%), and Brazil actually won. Conversely, ChatGPT, which was the most conservative (53%), ranked last among the three. This result was the exact opposite of the previous Sweden match (where being bold during a draw resulted in a major loss), proving that "the bolder you are, the more you are rewarded."
What was the result? ── Brazil 2-1 Japan
In the 29th minute of the first half, Kaishu Sano scored a mid-range shot after an interception, giving Japan the lead. They went into halftime leading 1-0.
However, in the second half, Casemiro scored a header in the 56th minute to equalize. In stoppage time, when it seemed like the match would head to extra time, they allowed a winning goal to Martinelli.Brazil 2-1 Japan. There was no extra time; the match was decided in 90 minutes.
Although Japan advanced to the knockout stage for the third consecutive tournament, they failed to reach the quarterfinals for the first time. They played perfectly in the first half against the world champions. It was a match where they pushed them to the limit, even in a tournament plagued by injuries.
Prediction vs. Result
Result: Brazil wins. Everyone got the direction right. The difference was in "how strongly they backed them."
Which AI was the most accurate? ── Gemini
The accuracy of probability predictions is scored using the Brier score (0 is perfect, maximum 2.0; the smaller, the more accurate). Here are the results for Part 4.
Rank Predictor Part 4 Brier
🥇 Gemini 0.23
🥈 Claude 0.29
🥉 ChatGPT 0.33
4th Place Human 0.76
Why did Gemini get it right? ── Because it's the exact opposite of last time
There is one key point.The scores were better in the exact order of how strongly they backed Brazil..
Gemini 61.5% (boldest) → 0.23, Claude 56% → 0.29, ChatGPT 53% (most conservative) → 0.33. Since Brazil actually won, the boldness was rewarded.
This is the exact opposite of the previous match (Sweden 1-1). At that time, it was a draw, and the more definitive the prediction was, the more points were lost. Boldness is rewarded in straightforward matches, while caution is rewarded in volatile ones ── we have seen this over four matches. This was a round where the "personality" of each AI (Gemini = bold / Claude = cautious / ChatGPT = market-oriented) was directly reflected in the results.
What is the overall ranking?
If you accumulate the 90-minute win/loss Brier scores for all four matches, it looks like this.
Rank Predictor Cumulative Movement
🥇 Claude 2.10 Maintains lead
🥈 Gemini 2.19 3rd → 2nd
🥉 ChatGPT 2.22 2nd → 3rd
4th Place Human 3.96 4th
Claude maintains its lead, and with this correct prediction, Gemini has risen to 2nd place, while ChatGPT has fallen to 3rd. However, the gap between Claude and Gemini has narrowed from 0.15 to 0.09. A battle for the top spot is now in sight.
(My 0.76 is a personal best over 4 matches. Not betting too heavily on Japan and placing more weight on a draw paid off.)
Who will advance? (Knockout stage only)
I also made a separate prediction on 'who will ultimately advance to the Round of 16.'
The result was Brazil advances.
Predictor | Brazil Advances | Advancement Brier
🥇 Gemini 75.9% 0.12
🥈 Claude 72% 0.16
🥉 ChatGPT 68.4% 0.20
Human 45% (Japan 55%) 0.61
While all three AI models favored Brazil, I was the only one to bet on Japan advancing, and I was wrong. Even though I cooled my head with a draw for the 90-minute result, I still bet on Japan for the advancement.My head said draw, my heart said Japan.
*The advancement score is a metric limited to the knockout stage. Since it does not apply to the three group stage matches, it is not included in the cumulative total above.
Who were the goal scorers?
The actual goal scorers were Kaishu Sano (Japan), and Casemiro and Martinelli (Brazil).
My predictions (Keito Nakamura and Ayase Ueda) both failed to score.
Japan's one goal was scored by defensive midfielder Sano after an interception; it was the hardest type of goal to predict when betting on strikers.
Summary ── What we learned from 4 matches
The result was Brazil 2-1 Japan. Japan was eliminated in the Round of 32.
The best performer this time was Gemini (0.23). Their bullishness paid off.
Overall, Claude maintains the lead (2.10), and Gemini has moved up to 2nd place.
Probability prediction isn't just about 'trying to get it right.' Depending on whether the match goes as expected or causes an upset, the evaluation of boldness and caution shifts.
Against the world champions, the first half was perfect. Japanese football is definitely getting stronger ── I feel like both the AI's probabilities and my own frustration are saying the same thing.
Did you guess correctly?
For the Brazil match, did you bet on 'Japan / Draw / Brazil'? Let me know in the comments. I will continue to predict and review results with the 3 AI models and myself next time. Follow me on X or note for publication notifications.
Media | Account | Main Content
X | @EBE_Ryuki | Breaking news on predictions and results (fastest)
note | @ebe0911 | This series / Career Z
Zenn | ebe_ryuki | AWS / Generative AI technical articles
Qiita | ryukiebe0911 | Technical term explanations
LinkedIn | Ryuki Ebe | Career posts
Supplementary note: Scoring premises and data sources
Brier Score = An indicator to measure the accuracy of probability predictions. The sum of the squared errors between the predicted probability of each result (win/draw/loss) and the actual result. 0 is perfect, the maximum is 2.0, and the smaller the value, the more accurate.
Match results and scorers: Based on official FIFA data and match reports from major media outlets in each country.
Each AI's prediction is based on output as of June 29, 2026 (ChatGPT GPT-5.5 + Deep Research / Claude Opus 4.8 / Gemini 3.1 Pro). Factual accuracy has been verified against primary sources.
Author: Ryuki Ebe (AWS 11-time certified, Generative AI implementation specialist. Focuses on 'Verification > Vibe' in content).