The video compares two AI trading models, Claude Code Opus 4.7 and Codex 5.5, in a one-hour perpetual trading challenge on Hyperliquid, with Codex 5.5 outperforming Claude by actively managing multiple trades and taking profits. Despite some technical issues, the creator praises Codex’s superior mathematical trading abilities and plans to continue refining AI trading strategies in future videos.
In this video, the creator continues a trading challenge series by pitting two AI models, Claude Code Opus 4.7 and Codex 5.5, against each other on the Hyperliquid trading platform. Unlike the previous challenge on Polymarket, this time the focus is on perpetual (perp) trades involving various assets such as commodities, stocks, and indices like Brent oil, SP500, Tesla, and Nvidia. The challenge rules are similar to before: each model is given a $100 budget and one hour to trade, with the goal of maximizing the dollar amount by the end of the period.
The setup involves giving both models the same prompt and framework, allowing them to research and plan their trading strategies within 15 minutes before executing trades in real-time. The creator built a custom dashboard to monitor the trades and countdown the hour-long challenge. Claude Code Opus 4.7 starts by shorting Brent oil and sets up multiple trades with reasons and confidence levels, while the creator tracks the live positions and updates during the challenge.
After Claude Code completes its trading hour, the same process is repeated for Codex 5.5, with both models running their strategies an hour apart due to account limitations. The creator then compares the results side-by-side, noting that Codex 5.5 outperformed Claude Code Opus 4.7 by a significant margin. Codex demonstrated a more active trading style, frequently adjusting positions and taking profits, whereas Claude mostly held a single short position on the SP500 throughout the hour.
The video highlights some technical issues encountered, such as API restrictions that affected Claude Code’s operations, but overall the creator expresses satisfaction with Codex’s performance. Codex has now won both the Polymarket and Hyperliquid challenges, prompting the creator to upgrade to Codex’s maximum subscription tier while downgrading Claude Code. The creator praises Codex’s strength in handling complex mathematical trading tasks despite Opus having a more advanced front-end interface.
In conclusion, the video serves as a comparative test of AI trading agents on a real trading platform, showing Codex 5.5 as the current leader in this type of challenge. The creator plans to continue refining strategies and conducting further tests, promising future videos that will delve deeper into the specific trading approaches used. Viewers are encouraged to like the video if they enjoy this content and to stay tuned for more updates on AI-driven trading experiments.