Claude Opus 4.8 Agentic Trading Performance Test
All About AIgo watch the original →
the gist
Claude Opus 4.8 was tested as an agentic trading bot on Hyperliquid and Polymarket, showing mixed results and significant stability issues compared to previous models.
Agentic Trading Performance
The author tested Claude Opus 4.8 as an autonomous trading agent over a one-hour period on Hyperliquid and Polymarket. The agent was configured with a heartbeat monitor set to pulse every 60 seconds to perform research and adjust positions dynamically. On Polymarket, the agent targeted 5-minute BTC trades by favoring trends that moved sufficiently from the opening window. On Hyperliquid, the agent attempted to ride news-confirmed trends, specifically selecting a long position on MU (Micron Technology) and silver.
Operational Stability and Results
The test yielded inconsistent performance and significant operational friction. While the Polymarket agent achieved a gain of 9.22, the Hyperliquid agent resulted in a loss of 5.6. The Hyperliquid agent performed well on ARM trades but suffered heavy losses on Samsung positions, which accounted for the majority of the deficit. Beyond the financial outcome, the author reported that Claude Opus 4.8 frequently halted execution despite explicit instructions to run for a full 60-minute duration. This lack of reliability led the author to conclude that Claude Opus 4.8 is currently less effective for agentic trading workflows than previous iterations, specifically citing CodeX 5.5 as a more stable and flexible alternative for managing heartbeat-based monitoring tasks.