The global football spectacle has collided with the artificial intelligence revolution, setting the stage for a high-stakes contest between algorithmic calculation and human intuition.

On June 24th, the first Chinese reality show to feature deep AI model participation in a World Cup broadcast, "Man vs. Machine: Who is the World Cup Prophet," will premiere on Migu Video. Co-produced by Lenovo Group and Migu, the show features celebrities like Su Xing, Zhan Jun, and Han Qiaosheng pitting their wits against 12 major AI models to predict match outcomes and scores.

However, a real-world "prediction battle" has already been raging for 13 days, covering the entire tournament and attracting over ten million participants.

Since the tournament's opening, the Tianxi AI Super Agent from Lenovo has acted as the convener, assembling 11 major AI models including DeepSeek, Kimi, Baidu's Wenxin, Qwen, and China Mobile's Jiutian. Together, they form 12 "AI teams" making predictions for all 104 matches, competing in real-time against human fans.

This is not only the world's first World Cup prediction contest featuring a collective of AI models against the public but also the first real-world test for China's major AI models on a global sporting stage. For every participating model, this is an open exam where the answers cannot be known in advance.

AI's Comeback: How Models Overtook Humans in 13 Days

Initially, AI did not receive applause. The first round of group matches was full of upsets. The most telling case was between Spain and Cape Verde. Before the match, none of the 12 AI models predicted a draw; 11 forecast a Spanish victory, and one placed a contrary bet on Cape Verde. However, all models assumed the game would see goals and a decisive result.

When the final whistle blew, the score was 0-0.

On the same day, the Iran vs. New Zealand match saw the AI camp miss the mark again. In a seemingly predictable match, all 12 models reached a rare unanimous consensus—predicting an Iranian win. The final result was a 2-2 draw, with none of the 12 AI models getting it right.

Across four matches that day, the 12 AI models made 48 predictions on match outcomes, with only one correct call. At that point, the AI prediction accuracy rate was just 35%, significantly trailing the human camp.

However, a turning point soon emerged. As the second round of group matches entered a phase where stronger teams asserted dominance, with expected results like the USA's 2-0 win over Australia and the Netherlands' 5-1 victory over Sweden materializing, the prediction accuracy of the AI camp began to climb sharply. Models like China Mobile's Jiutian, Baidu's Wenxin, DeepSeek, Qwen, and Lenovo's Tianxi AI achieved consecutive correct predictions, rapidly boosting the overall hit rate. By June 24th, the overall accuracy of the 12 "AI teams" had risen to 57%, surpassing the human participants' overall accuracy of 52.5%.

A 4.5-percentage-point gap may seem small, but recent trends in win rates show a clear divergence in the direction of the two curves—the AI line continues to climb, while the human line has largely flattened.

In terms of model rankings, based on prediction data from 46 matches, China Mobile's Jiutian currently leads with a 63% prediction success rate. Models like Lenovo's Tianxi AI, Qwen, and Tencent's Hunyuan form a second tier with a 60.9% success rate.

More importantly, no AI company had the opportunity to know this ranking in advance, nor could anyone design it beforehand. World Cup results are decided by the 22 players on the pitch, not by the models. It is precisely for this reason that this data holds special value.

A New Benchmark: The Significance of This Data Set

To understand the value of this experiment, one must first understand how China's AI industry has validated its capabilities in recent years.

For a long time, evaluating large model capabilities has primarily relied on three methods: benchmark tests, product data, and event marketing. However, all three have inherent limitations. Benchmark tests occur in a lab environment, far removed from the complexity of the real world. Product data is held by individual companies, making cross-comparison difficult. Event marketing can generate buzz, but buzz does not equate to capability. The common problem with these three approaches is that the way conclusions are generated can be designed, and the credibility of designed conclusions is inherently discounted.

The World Cup provides a fundamentally different validation framework.

Before each match kicks off, the 12 major AI models must present their judgments under the same set of rules. The outcome is decided by 22 players on the field, beyond the control of any AI company. Once a prediction is made public, it cannot be altered after the fact; once the result is final, it provides instant verification. This mechanism, applied across 104 matches, produces a sample of capabilities tested match-by-match in the real world, not extrapolated numbers from a lab.

The data accumulated over 13 days already reveals a clear pattern: AI excels at "orderly" questions but struggles with "trap" questions. When the strength hierarchy is clear and matches unfold according to form, AI's prediction probability is very high. When football enters moments dominated by draws, upsets, on-field fluctuations, and emotional variables, AI quickly loses its grip. This conclusion isn't stated by any single participating model; it is presented match by match through the outcomes of 104 games.

For China's AI industry, the value of this data lies in its "immutability"—it was generated under conditions of public scrutiny, real-time verification, and unalterable results. This is almost the first instance of such capability validation in the history of domestic AI.

What makes the data even more meaningful is the participation base of over ten million people. The human prediction camp's 52.5% accuracy constitutes a real, sizeable comparative baseline. The AI is not winning against a hypothetical opponent but against a real judgment sample exceeding ten million instances.

The Platform Builder: Lenovo's Role

The existence of this validation framework hinges on one prerequisite: someone must have access to the World Cup "ticket."

Lenovo Group is the Official Technology Partner for the 2026 FIFA World Cup, deeply involved in building the core technological framework for this tournament with end-to-end, full-domain AI technology. This role is not merely a marketing label but represents genuine, deep technical integration.

For this World Cup, Lenovo has deployed the FIFA AI Pro World Cup Football AI Super Agent, providing tactical analysis support for all 48 participating teams. Lenovo's 3D digital human visualization solution has increased offside determination accuracy to a "scalp-level" precision, creating digital avatars for all 1,263 players to aid in the visual presentation of the tournament's semi-automated offside technology. Furthermore, the referee perspective AI video enhancement system, developed in-house by Lenovo in under a year, has for the first time stably integrated the referee's first-person view into the global broadcast, presenting clear footage to a worldwide audience.

Additionally, Lenovo is involved in operating core hubs like the Dallas International Broadcast Centre, the Miami Event Operations Centre, and the Miami Technical Command Centre, ensuring the real-time operation of the tournament across 16 cities in three countries.

This capacity for deep involvement in World Cup operations is not a path any AI vendor can replicate alone, granting Lenovo the unique conditions to establish this World Cup prediction testing ground.

Building on this, the Tianxi AI from Lenovo initiated the "Man vs. Machine" battle in the role of "convener," gathering 11 major domestic AI models to present their predictions on the same stage, with match results as the sole criterion for judgment. This design inherently ensures the experiment's credibility, as no single participant can control the outcome.

Ultimately, the "World Cup Prediction: Man vs. Machine Battle" has created the world's first World Cup prediction contest featuring a collective of AI models against the public, attracting over 18 million users to participate. This scale has transformed it from a brand activity into a publicly significant statistical experiment.

Beyond a Single Beneficiary

Considering these three aspects together reveals the true value of Lenovo's strategic move during the 2026 World Cup.

It is more than just high-profile brand marketing. Through the "Man vs. Machine" battle, Lenovo has built the first public platform for capability validation in China's AI industry that operates continuously in the real world, is visible to the public, and prohibits post-facto modifications. On this platform, the actual performance of various large models is recorded match by match, and the boundaries of AI capabilities are gradually laid bare for all to see. This "referee's platform" provides China's AI industry with a public capability benchmark it has never had before.

For Lenovo, it occupies the role of builder and operator of this coordinate system—not as a large model itself, but as a platform entity that can bring major models to the same test paper and let the real world be the judge.

This role holds significant strategic value in the AI industry's transition from the "hundred-model battle" phase to real-world application. When major models are all seeking more real-world validation, a platform that can provide such validation opportunities is itself a scarce piece of infrastructure.

The premiere of the "Man vs. Machine: Who is the World Cup Prophet" show at 21:00 on June 24th marks the point where this experiment upgrades from "data-readable" to "process-watchable." The show's 20 live broadcasts, public predictions by guests alongside AI, and real-time post-match analysis will extend the influence of this "referee's platform" from millions of participants to an even larger audience, marking the beginning of the experiment entering a grander stage.

As of now, the "Man vs. Machine" battle is still ongoing, with data from subsequent matches continuing to be generated. The complete conclusions of this experiment will only be finalized after all 104 matches are played. But one thing is already clear: in 2026, this "AI World Cup inaugural year," Lenovo has chosen a more substantial way than slogans to prove the value of AI—letting real match results speak for Chinese AI.