Descriptions:
The All About AI channel puts Claude Opus 4.8 through a live one-hour agentic trading session across two platforms — Hyperliquid (perpetual futures) and Polymarket (5-minute BTC prediction markets) — using Claude Code at high-effort mode with the same prompts as a prior Opus 4.7 run to allow direct comparison. The agent autonomously selects its trading strategy, manages position sizing, and adjusts in real time via a heartbeat daemon that polls every 60 seconds.
Results were mixed: Polymarket returned +9.22% over the hour, improving on the previous run, while Hyperliquid came in at -5.6% — worse than the 4.7 baseline. The loss on Hyperliquid traced largely to three consecutive losing long positions in Samsung, which accounted for roughly $9 of the $15 total loss. Long positions in ARM performed well, ending positive in both directions.
The host is upfront that a single one-hour snapshot is not a statistically valid benchmark and notes that a longer continuous evaluation is underway. Still, the video serves as one of the more concrete real-money demonstrations of Opus 4.8’s autonomous decision-making, including how the model narrates its own reasoning when asked to explain strategy before trading begins. Viewers interested in agentic finance applications will find the live session footage and dashboard monitoring useful context for evaluating the model’s behavior in an unstructured, real-stakes environment.
📺 Source: All About AI · Published May 29, 2026
🏷️ Format: Benchmark Test
