Descriptions:
This video sets up a direct head-to-head challenge between two leading AI coding agents — Claude Code running on Opus 4.7 and OpenAI’s Codex 5.5 — competing on the Hyperliquid perpetuals trading platform. Each agent is given a $100 budget and 60 minutes to research, plan, and execute trades on XYZ perpetual contracts spanning commodities (Brent oil, natural gas, silver, gold), equity perps (MRVL, Hood), and currency instruments. Both run in maximum-effort modes — Claude on ‘high effort’ and Codex in ‘YOLO mode’ — with identical prompts instructing them to browse the web for data, build a strategy, and actively monitor and adjust positions via web socket connections throughout the hour.
The creator built a custom real-time dashboard to track positions, PnL, and agent activity side-by-side. Claude Code’s opening move is a short on Brent oil alongside positions in MRVL and Hood, with reasoning provided for each trade. The setup follows a prior Polymarket trading challenge in which Codex outperformed, with the creator reporting that the evolved Codex strategy had run profitably for 48 consecutive hours before this Hyperliquid test.
The challenge offers a rare look at autonomous financial decision-making under real market conditions and time pressure, revealing how each agent handles strategy formation, tool orchestration, and dynamic position management — capabilities increasingly relevant as agentic AI moves into consequential real-world tasks.
📺 Source: All About AI · Published May 28, 2026
🏷️ Format: Benchmark Test
