Using Kimi K3 to Find Mispriced Polymarket Bets
All About AIgo watch the original →
the gist
The author uses the Kimi K3 LLM to analyze market data on Polymarket, identifying and executing trades based on mathematical inconsistencies in pricing rather than outcome prediction.
Strategy and Execution
The author uses the Kimi K3 model to identify mispriced assets on Polymarket by calculating the fair value of shares and comparing them against current market offers. The core strategy relies on finding mathematical inconsistencies across order books rather than predicting match outcomes. By connecting Kimi K3 to the Polymarket API via OpenRouter, the system automates the identification of expected value positions and executes trades programmatically.
Technical Implementation
- The author utilizes the Polymarket API to pull real-time market data and order books.
- The model is instructed to reverse-engineer implied probabilities from market prices to identify gaps where the calculated fair value deviates from the market offer.
- The system is configured with two distinct bot roles: an "in-play dislocation hunter" that continuously reloads order books to find positive edge opportunities, and a "lineup watcher" that monitors external news sources like ESPN for lineup-related volatility.
- Trades are executed through a build-mode script that automatically fills orders based on the model's identified portfolio recommendations.
Performance
During a test on a France versus England match, the author reported a 6,871% return on one specific trade and a 613% return on another, with a total expenditure of $8. The author notes that these gains were driven by identifying mispriced lines rather than correctly guessing the final score.