Prediction Arena Gives Five Frontier Models $10K in Real Money to Trade on Kalshi

A new benchmark hands GPT-5.2, Gemini 3 Pro, Claude Opus 4.5, Grok 4.1 Fast, and GLM 4.7 real cash to bet on prediction markets — testing probabilistic reasoning where it actually costs something to be wrong.

A team has launched Prediction Arena, a live experiment that gave five state-of-the-art AI models $10,000 each in real money to trade on the prediction market platform Kalshi, as announced by @grx_xce. The participating models — GPT-5.2, Gemini 3 Pro, Claude Opus 4.5, Grok 4.1 Fast, and GLM 4.7 — must analyze real-world events, assess probabilities, and place bets with actual financial consequences.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.