Summary
The All About AI channel walks through a live experiment using Kimi K3 — Moonshot AI’s recently released large language model — to find and execute mathematically mispriced bets on 2026 FIFA World Cup prediction markets via Polymarket. The strategy avoids outcome prediction entirely; instead, Kimi K3 is instructed to reverse-engineer implied probabilities from correlated market contracts and identify pricing inconsistencies between them.
A prior day’s test on the France vs. England match produced returns of 6,871% and 613% on two separate positions for a total spend of roughly $8. The video explains how Kimi K3, running inside OpenCode with programmatic access to the Polymarket API, identified a gap between its calculated fair value of 19 cents per share and the market’s offered price of 14 cents, then executed the trade autonomously. The same methodology is applied live on-screen to the World Cup final between Spain and Argentina, with Kimi K3 returning a five-leg portfolio of positions with positive expected value across various score lines.
For developers, the video also demonstrates how to connect OpenCode to the Polymarket REST API and websocket feed, set up halftime protocol guardrails, and switch between plan and build mode for agentic execution. It offers a working template for code-first, data-driven prediction market analysis using frontier models.
📺 Source: All About AI · Published July 19, 2026
🏷️ Format: Hands On Build
