- Alibaba’s AI assistant uses player data, weather forecasts and local geography to predict match outcomes
- Users who outperform the model can compete for cash prizes, while participation helps fund football pitches in rural schools
In the 2010 South Africa World Cup, an octopus named Paul gained fame by correctly predicting all eight matches, becoming a legend in World Cup forecasting.
Sixteen years later, AI has taken up Paul’s mantle—not by picking flags out of a box, but by analyzing vast amounts of data to predict outcomes.
As the 2026 FIFA World Cup approaches, Alibaba’s Qwen has entered the increasingly crowded sports prediction arena with a new AI-powered football forecasting assistant, betting that data-driven analysis can hold its own against human intuition.
The tool, which has been opened to all users ahead of the tournament, marks one of the first large-scale attempts to bring generative AI into World Cup prediction contests.
The launch comes as fans debate whether large language models can outperform seasoned football pundits in forecasting match results.
Cheng Fei, head of Qwen’s football prediction product, said the system draws on a wide range of data sources, including historical head-to-head records, player valuations, injuries and squad information.
Geographical, weather, environmental factors
For the first time, the model also incorporates geographical data from the United States, Canada and Mexico, as well as weather forecasts during the tournament.
As an example, Cheng pointed to Norway’s June 22 group-stage match against Senegal. Despite Norway’s apparent advantage on paper, led by star striker Erling Haaland, Qwen predicts a 1-1 draw.
“The reason for that prediction is the weather,” Cheng said. “Haaland has spent most of his career in the cooler climate of Northern Europe. High temperatures in North America could significantly increase physical demands.”
The model also factors environmental conditions into its forecast for the opening match between host nation Mexico and South Africa.
Qwen predicts a Mexican victory, citing both home advantage and Mexico City’s altitude of more than 2,200 meters, which could affect visiting players.
Limitations of AI
Asked whether AI can accurately predict football matches, Cheng acknowledged the limitations of the technology.
“We’ve optimized the model as much as possible, but football’s greatest appeal is its unpredictability,” he said. “If someone tells you they can predict match results with 100% accuracy, they’re probably not an AI—they’re a fraud.”
Alongside the prediction service, Qwen launched a football philanthropy campaign linking user participation to grassroots sports development.
Supporting rural football
Under the initiative, accumulated points generated through AI prediction games will trigger donations to build or renovate football pitches at schools in rural and underserved areas, with a target of supporting at least 50 schools.
“Years from now, nobody may remember how many matches Qwen predicted correctly,” Cheng said. “But if this campaign results in 50 more football pitches and inspires more children to fall in love with the sport, that will matter much more.
“Chinese football needs professional leagues and youth development systems, but it also needs a grassroots foundation,” Cheng added.
The promotion also includes incentives for users. Participants who make predictions for more than 80 of the tournament’s 104 matches and achieve a higher accuracy rate than Qwen will qualify for a drawing for a 10,000-yuan ($1,475) cash prize.
Users who submit predictions for more than 32 matches can enter a separate draw for one of 1,000 Qwen AI glasses.
