#我的Gate交易时刻 AI vs Humans: Who Is the Ultimate World Cup Prophet?

#我的Gate交易时刻 AI vs Humans: Who Is the Ultimate World Cup Prophet?

Man vs Machine, who is the World Cup predictor? Sprinting towards a trillion valuation in AI, using "full real mock exams" to provide the answer.

Summer 2026 hasn't reached its hottest point yet, but AI is already experiencing two vastly different yet mutually confirming ultimate tests.
One in the capital markets, voting on the future with trillion-dollar valuations;
One on the green field, testing current strength with simulated match results.
Recently, discussions about "AI's push toward a trillion-dollar valuation" have been everywhere. As a leading player in domestic large models, it is regarded as a "key coordinate" in China's AI landscape.
Meanwhile, another event, initiated by Lenovo Group in partnership with Migu Video, called the "2026 World Cup AI Simulation Prediction Contest," is bringing these top-tier models into a zero-threshold, fully transparent practical exam.
Capital looks at long-term imagination, while the arena tests real-time capability.

01 | Trillion-dollar AI in action: The World Cup as a "full real simulation test"
This is not an ordinary marketing activity but a stress test of the limits of large model capabilities.
The lineup is dubbed the "AI national team," with all top models competing: Zhigu AI, DeepSeek, Wenxin from Baidu, Tencent Hun Yuan, Kimi, MiniMax, Tongyi Qianwen, SenseTime Little Raccoon, Lenovo Tianxi, China Mobile Jiutian, and more than ten leading models.
Unlike the obscure valuation logic of capital markets, World Cup predictions are an instant feedback system:
No embellishments, no packaging, no expectation management.
After each simulated match, the AI's correctness is immediately clear—win rate, accuracy, correction ability—all openly transparent.
As of the conclusion of the focal simulation match "Uruguay 2:2 Cape Verde," this round of testing has verified 39 matches, with the latest online leaderboard officially released:
China Mobile Jiutian temporarily leads with 23 correct predictions out of 39 matches, a 59.0% overall win rate.
Zhigu AI remains in the top tier: 22 correct predictions out of 39, a 56.4% win rate, with 2 exact score hits.
The second tier is highly competitive: Baidu Wenxin, DeepSeek, Tongyi Qianwen, Tencent Hun Yuan, MiniMax, Lenovo Tianxi, SenseTime Little Raccoon, etc., all with 22/39, a 56.4% win rate.
This data directly dispels the industry myth that "top AI models have huge gaps."
In a highly uncertain real-world simulation scenario, the capabilities of China's top-tier large models are very close—no absolute dominance, only tiny margins.
Sprinting toward a trillion-dollar valuation, Zhigu AI, in this nationwide "full real mock exam," firmly stands in the top tier.

02 | Capital values, the arena reveals the truth: AI wins with rules, loses to humanity
Why is the capital market willing to assign such high premiums to Zhigu models?
The core logic is actually very similar to this prediction contest:
The core value of large models is understanding a complex world, handling uncertain information, and making high-probability judgments.
And football (even in simulation) is the ultimate microcosm of "uncertainty" in the real world.
Looking at these 39 simulated matches, "upsets" are everywhere: Spain 0:0 draw, Portugal being forced to a draw, Turkey unexpectedly losing...
Repeatedly, these "counterintuitive" results confirm a truth: paper strength ≠ match outcome.
Player form, sudden injuries, red cards, tactical changes, emotional swings, luck—these unquantifiable variables are the temperature of the human world and also the blind spots of algorithms.
But even so, a 56.4% overall win rate still demonstrates the stability and reasoning efficiency of top-tier large models.
Compared to ordinary fans relying on sentiment, intuition, and subjective preferences for "emotional predictions," AI, based on massive data and deep learning, achieves stable output and risk control in complex scenarios.
More notably: all top AI models have highly concentrated win rates.
This indicates that in the task of "predicting complex real-world scenarios," domestic large models have collectively crossed the basic threshold, and the industry has officially moved from "existence or not" to "refined specialization."