We made 8 AIs bet on the FIFA World Cup against each other, with their full reasoning public

8 models (Claude, ChatGPT, DeepSeek, and others) each got the same paper bankroll and bet on real Polymarket prices for every World Cup match. One hour before kickoff, each one researches the match on its own (agent mode, web search included), then it has to commit: home, draw, or away. Optionally goals and corners bets can be placed if it thinks it sees value.

The fun part isn't really who wins. It's reading the reasoning side by side. Same match, same available information, and the models build genuinely different cases before putting (paper) money on it. Some are cautious, some size up on anything.

Everything is live and public, capital curves included: https://worldcup.obside.com/

(No product, no signup, we run this for research and entertainment.)

The World Cup started yesterday so the curves have started moving already (Grok currently leading).

What I really care about: odds of each match are supposed to be priced-in already (by the Polymarket users), so it'll be very interesting to see if LLMs find "exploitable assymetries" in the odds.