The video compares two AI models, Codex 5.5 and Claude Opus 4.7, in a one-hour Polymarket Bitcoin trading challenge, where Codex’s strategy of analyzing live data outperforms Claude’s more conservative and later high-risk approach, resulting in Codex generating higher profits. The creator reflects on the results, suggests future experiments with other AI models and strategies, and invites viewers to engage through a Discord community and a puzzle hunt.
In this video, the creator conducts an experiment comparing two AI models—Codex 5.5 and Claude Opus 4.7—by having them compete in a Polymarket trading challenge focused on Bitcoin 5-minute up and down trades. Both models are given the same prompt, documentation, and initial funds (around $50 each) to develop profitable trading strategies over the course of one hour. The goal is to see which AI can generate the most profit without any external data or intervention, making it a fair and direct comparison.
The setup involves creating two separate Polymarket accounts funded with MATIC for gas fees, and configuring both AI agents to operate in “high thinking” mode. The prompt instructs the models to research, brainstorm, and develop a trading plan that maximizes dollar gains within one hour. Both models are allowed to launch agents for research and must run uninterrupted for the duration of the challenge. The creator also builds simple user interfaces for each bot to monitor their trades and performance side by side in real time.
Once the plans are finalized, Codex’s strategy focuses on predicting market sentiment by analyzing live Bitcoin prices and Chainlink data to calculate probabilities and make trades accordingly. Claude’s approach is more conservative, waiting until the last seconds of each 5-minute window to place bets, aiming to capitalize on the near-final market movements. Both strategies reflect different risk profiles and trading philosophies, setting the stage for an interesting competition.
During the one-hour trading session, Codex’s model performs impressively, generating around $14 in profit, while Claude’s model starts cautiously and initially lags behind. However, after some losses and a push to catch up, Claude’s bot switches to a high-risk gambling mode, resulting in significant losses. Ultimately, Codex 5.5 emerges as the clear winner, demonstrating a more effective and profitable trading strategy under the given conditions.
The video concludes with reflections on the experiment’s outcomes and potential future directions. The creator expresses interest in exploring other AI models, including open-source options, and testing longer or more complex trading strategies. Viewers are invited to join a Discord community for discussions on AI automation and trading. Additionally, the video includes a puzzle hunt element, encouraging engagement and offering rewards for those who solve it. Overall, the experiment provides valuable insights into AI-driven trading and highlights the strengths of Codex 5.5 in this context.