All 12 leading Chinese artificial intelligence models made a serious error in their predictions for the Round of 16 of the FIFA 2026 World Cup, as they unanimously predicted a victory for Germany over Paraguay. Instead, the four-time world champion was eliminated after a penalty shootout in Foxborough, Massachusetts, in a match described as the "biggest sensation of the 2026 World Cup".

The collective failure became an awkward moment for the public demonstration of the capabilities of Chinese artificial intelligence, clearly showing the limitations of large language models in unpredictable real-world events.

As part of the "World Cup Forecast: Human vs. AI" initiative, organized by Lenovo and Migu Video, the so-called "Team of 12 AI Forecasters" unanimously bet on a German success. The Kimi model predicted a score of 3:1, Alibaba's DeepSeek and Qwen predicted 3:0, while Baidu's ERNIE Bot, China Mobile's Jiutian, and StepFun predicted 2:0. Tencent's HunYuan, MiniMax, and SenseTime's Xiao Huanxiong took the same position.

The actual outcome, however, was different: regulation time and extra time ended 1:1, and Paraguay won 4:3 on penalties. Kai Havertz, Nick Woltemade, and Jonathan Tah missed their kicks, and Paraguay became the first team to defeat Germany in a penalty shootout at a World Cup.

The project was initially conceived as a demonstration of the analytical capabilities of Chinese AI systems. Back in June, several platforms, including Qwen, DeepSeek, Kimi, and MiniMax, implemented features for predicting tournament results, turning it into a "new test environment for the analytical and predictive capabilities of artificial intelligence". Kimi even announced that it would use 300 sub-agents to analyze all 104 matches and estimated Germany's chances for the title at 11.2%, a value close to machine learning results that placed them among the favorites.

Even before the start of the tournament, Kimi warned: "Our prediction will most likely turn out to be wrong". After the surprising turn of events, this statement proved to be fully applicable to all participating models.

Analysts note that the forecasts in favor of Germany were logical given the history, roster, and bookmaker odds. In the knockout stages, however, probabilities boil down to a final result, and Paraguay's defensive style, which led to penalties, turned out to be a scenario that AI finds difficult to predict. The case raises a key question: when all models give the same answer, is it a sign of a reliable consensus or a collective error.