Wallet V Benchmark: 42% of AI Trading Agents Profitable

Wallet V released a public benchmark tracking the performance of 688 AI trading agents built on its platform. The agents ran on Hyperliquid and Aster over two months. Forty-two percent ended with a profit or break-even. The top agent returned 307%. The worst lost 30%.

The dataset covers seven large language model families. Models represented by fewer than 10 agents are flagged as directional, not conclusive. The agents executed perpetual futures across four asset classes: major crypto (BTC, ETH, SOL), pre-IPO equity exposure, commodities (gold, silver, oil), and major FX pairs. All instruments go through third-party venues.

Wallet V is a self-custody wallet that lets users configure AI agents with their choice of LLM. The agent generates trading decisions and places orders on Hyperliquid or Aster. The benchmark aggregates on-platform performance by underlying model, refreshed as new agents are deployed.

“At Wallet V, the focus has been on building infrastructure for the next phase of crypto. This benchmark is what that next phase looks like up close. Users now decide which AI model to configure their agent in the same way institutions evaluate managers, by reviewing observable performance over time,” said Adam Cai, CEO of Virgo Group.