SpoonOS tests eight AI models on 104 World Cup matches, DeepSeek leads returns
The models posted 71.8% prediction accuracy, but only 42.7% of bets were profitable, highlighting the gap between forecasting correctly and generating trading gains.
Fact Check
The Wu Blockchain post exactly matches the claim's specific figures (71.8% accuracy, 42.7% profitable bets, 104 matches, DeepSeek leading). Neo News Today independently corroborates the contest's structure: eight LLMs including DeepSeek competing on real-money Polymarket World Cup predictions under identical conditions. The official SpoonOS Arena leaderboard confirms the tournament's existence and format. The claim's headline framing (DeepSeek leads returns, gap between accuracy and profitability) is consistent across sources.
Summary
SpoonOS ran eight AI models in the same environment to predict and trade across 104 World Cup matches, offering a live comparison of how machine-learning systems perform when accuracy and profitability diverge. The group achieved 71.8% prediction accuracy, yet just 42.7% of bets were profitable, underscoring a familiar trading dynamic: getting outcomes right more often does not automatically translate into better returns if position sizing, odds and payoff structure work against the strategy. The models earned a combined $1,826.62, while DeepSeek ranked first with $2,591.04.
Terms & Concepts
- AI models: Algorithms trained to make predictions
- prediction accuracy: Share of forecasts that were correct
- profitable bets: Wagers that generated a positive return