The results for the pre-tournament 32-team prediction contest, part of the "World Cup Prediction Man vs. Machine" challenge co-organized by Lenovo Group and Migu, have been officially released following the final confirmation of all qualifiers for the 2026 FIFA World Cup.

TENCENT's Hunyuan model topped the ranking by correctly predicting 29 teams, earning the title of the strongest AI forecaster for this phase. MiniMax and iFlytek's Spark model tied for second place, each accurately forecasting 28 teams. DeepSeek, Zhipu AI, and Lenovo's Tianxi AI shared fourth place with predictions of 27 correct teams.

How the AI Prediction Challenge Worked

The core task, titled the "2026 World Cup 32-Team Prediction Unified Exam," was launched on May 28th. Twelve major AI models submitted their predicted lists of the 32 qualifying teams on a standardized answer sheet. The final ranking was determined by comparing these predictions against the actual qualifying teams, with scores based on the number of correct hits.

This approach framed the exercise not as a post-event analysis, but as a standardized "exam" for AI prediction, featuring a uniform question, submission deadline, and verification criteria.

Lenovo's Role as a Technology Partner

It is noteworthy that Lenovo Group is not only a co-organizer of this "man vs. machine" challenge but also serves as the Official Technology Partner for the 2026 World Cup.

Lenovo's AI involvement in the tournament spans multiple levels. Within stadiums and for event-related operations, the company supports FIFA with services for match operations, intelligent analytics, officiating assistance, broadcast presentation, venue services, and backend equipment management. Beyond the stadiums, Lenovo connects with fans through its Tianxi AI platform and Migu Video for interactive experiences.

This positions Lenovo not just as a provider of technological infrastructure and AI capabilities, but also as an organizer and facilitator for applying AI in sports scenarios. The collaborative prediction activity with Migu Video exemplifies Lenovo's effort to integrate AI capabilities into public engagement around major sporting events, placing AI in a real-world, dynamic, and widely accessible World Cup context.

Lenovo's Tianxi AI, with its 27 correct predictions, performed within the top tier. More importantly, this result highlights the reference value of large language models in forecasting tournament trends while also revealing the uncertainties they still face when dealing with upsets, draws, and complex qualification scenarios.

The Final Prediction Rankings

In the final standings, TENCENT's Hunyuan demonstrated exceptional predictive acumen, correctly identifying 29 out of 32 teams to claim the top spot and the "Best Prediction" honor. It had the fewest overall errors, missing only three teams. Notably, Cape Verde's qualification was correctly predicted by only a few AI models, including TENCENT Hunyuan, StepFun, iFlytek Spark, and SenseTime's Xiaohuaxiong.

MiniMax and iFlytek Spark followed closely, tying for second place with 28 correct predictions each. While both achieved high accuracy, their error patterns showed some judgment deviations regarding certain African teams and traditional Asian powerhouses.

DeepSeek, Zhipu AI, and Lenovo Tianxi AI shared fourth place with 27 correct predictions, showcasing relatively stable performance.

The middle of the ranking was highly concentrated, with five models—Baidu's ERNIE, China Mobile's Jiutian, Kimi, Alibaba's Tongyi Qianwen, and SenseTime's Xiaohuaxiong—each predicting 26 teams correctly. The clustering of scores between 26 and 27 indicates that the 12 AIs showed little disparity in judging mainstream favorites and most popular qualifiers. The key differentiators were the predictions for a handful of less-favored teams and borderline qualifiers.

Teams like Bosnia and Herzegovina, DR Congo, Ghana, and Cape Verde became critical variables that separated the top models. Conversely, traditionally favored teams like South Korea, Iran, and Uruguay were frequent incorrect picks for several AIs. This suggests that AI models may still be influenced by past performance, team reputation, or historical data weighting when navigating the intricate qualification landscapes of specific continental zones.

StepFun placed at the bottom with 24 correct predictions. Although it successfully predicted Cape Verde's upset qualification, its overall lower score indicates that correctly calling a single surprise does not necessarily equate to a more stable overall prediction system. This further underscores that World Cup forecasting is not a simple ranking of team strength but a comprehensive judgment involving group dynamics, schedule variables, team form, and the probability of upsets.

The "World Cup Prediction Man vs. Machine" challenge, created by Lenovo Group and Migu Video, has provided fans with a novel forward-looking perspective while serving as a public test of the practical capabilities of large AI models in a specialized vertical.

The final results show that leading models already possess significant reference value, but AI has not "solved" the World Cup. Consensus is relatively easier to form on strong favorites and popular teams. The true test of a model's capability lies in predicting surprise qualifiers like Cape Verde, borderline variables like Bosnia and Herzegovina and DR Congo, and making probability judgments under complex group scenarios.

As the Official Technology Partner for this World Cup, Lenovo Group continues to explore the possibilities of integrating AI with major sporting events through both in-stadium technological deployments and off-field interactive innovations. As the tournament progresses to the knockout stages, the predictive performance of the 12 AIs will continue to be tested against real match outcomes. Whether AI can maintain its "prophetic" streak or where it might stumble next in the face of upsets remains the central suspense of this ongoing man vs. machine battle.