‍#广场预测世界杯赢40000U

12 AI Large Models Release World Cup 32-Team Prediction Accuracy Results: Cape Verde and DR Congo Stand Out, 11 Incorrectly Predicted Uruguay

Before the World Cup knockout stage, 12 AI large models from China have submitted their report cards.

After all 32 teams for the 2026 FIFA World Cup in the United States, Canada, and Mexico were confirmed, the "World Cup Prediction Human vs. AI Battle" launched by Lenovo Group and Migu Video revealed the final prediction accuracy of these 12 AIs for the 32-team qualification results.

Based on the final confirmed list of 32 teams, Tencent Hunyuan correctly predicted 29 teams, ranking first; MiniMax and iFlytek Spark correctly predicted 28 teams, tied for second; DeepSeek, Zhipu, and Lenovo Tianxi AI correctly predicted 27 teams; Baidu Wenxin, CMCC 9tian, Kimi, Qianwen, and SenseTime Little Raccoon correctly predicted 26 teams; and Jieyue correctly predicted 24 teams.

The World Cup has plenty of data, but also plenty of uncertainty. Some data can be absorbed and calculated by models, such as team strength before matches, historical records, player conditions, odds changes, and point standings. However, other factors constantly disrupt model predictions, such as red cards, injuries, weather, last-minute substitutions, psychological fluctuations, and unexpected goals.

AI predicting football does not mean AI can accurately understand football. Football is a highly uncertain sport: strong teams can stumble, weak teams can cause upsets, and draws often defy model and fan intuition. Previously released phased data from Lenovo and Migu also showed that after the first round of group matches, AIs performed better in matches with clear strength disparities but were unstable in predicting draws, upsets, and matches involving certain Asian teams.

The highest overall accuracy was from Tencent Hunyuan, which correctly predicted 29 qualified teams—the only top score among all AIs and all human participants. It correctly answered most routine questions and successfully predicted the upset of Cape Verde, only missing Bosnia and Herzegovina, DR Congo, and Ghana.

What is the difference between predicting 29 qualified teams and predicting 24 qualified teams?

It’s not Spain, Germany, or Brazil—most participants correctly chose these popular teams. What truly made the difference were two teams with low pre-tournament support rates.

Cape Verde: Only 11.54% of users predicted it would qualify. Among the 12 AIs, only 4 predicted correctly: Tencent Hunyuan, iFlytek Spark, SenseTime Little Raccoon, and Jieyue Xingchen.

But Cape Verde ultimately became the biggest Cinderella story of this World Cup so far.

DR Congo: Only 18.25% of users predicted it would qualify. Among the 12 AIs, only Jieyue Xingchen predicted correctly.

In this 32-team prediction, the most typical high-consensus loss item was Uruguay.

91.63% of users predicted Uruguay would qualify, and 11 out of 12 AIs made the same judgment. As a four-time World Cup champion and traditional South American powerhouse, Uruguay received high recognition from both human users and the 12 AIs before the tournament. However, Uruguay ultimately failed to qualify for the 32-team tournament.

Similar cases include South Korea and Iran. South Korea received 88.46% user support, with 11 AIs predicting its qualification, but it ultimately failed to advance; Iran received 77.17% user support, with 10 AIs giving the same prediction, and it also failed to qualify.

In contrast, Jieyue Xingchen became the most distinctive sample among the 12 AIs. It was the only AI that did not choose South Korea, the only one that did not choose Uruguay, and the only one that chose DR Congo. It demonstrated a prediction path different from other AIs when it came to upsets.