Wallet V, a Web3 self-custody wallet incubated by Virgo Group, has published a public performance benchmark for AI trading agents running on the Hyperliquid and Aster platforms. The dataset covers 688 agents built by Wallet V users over the previous two months, offering a rare look at how autonomous models actually perform in live futures markets.
Mixed results across the agent cohort
According to the data, 42% of agents finished the period with a profit-and-loss balance of zero or higher. That means more than half of the models lost money. The spread between best and worst is wide: peak agent-level return on investment ranged from negative 30% on the lowest-performing model to positive 307% on the highest.
That kind of variance isn't surprising for an emerging field, but it underscores how uneven AI trading can be. The agents aren't just running on one kind of model — they span seven large language model families. Wallet V notes that models represented by fewer than 10 agents are reported as directional rather than statistically conclusive, a caveat that cautions against reading too much into small-sample results.
What the agents traded
The agents executed perpetual futures across four asset classes: major digital assets like Bitcoin, Ethereum, and Solana; pre-IPO equities; commodities including gold, silver, and oil; and major foreign exchange pairs. That mix shows the platform isn't limiting itself to crypto — it's targeting anything with a futures market and enough liquidity for automated trading.
Hyperliquid and Aster are relatively newer venues for this kind of activity, and Wallet V's decision to open up the benchmark suggests it wants to attract more developers and traders to test their own models against the published results.
Plans to broaden the scope
Wallet V says it plans to extend the benchmark to include newer model families, prediction markets, copilot trading analytics, and personalized AI prompt generation. Those additions would give users more ways to compare agent performance — and potentially more tools to build or improve their own models.
For now, the benchmark is a snapshot, not a verdict. The 42% profitability figure doesn't account for fees, slippage, or risk management, and the data covers a relatively short period. But it's one of the first public datasets to compare hundreds of AI agents side by side on the same infrastructure.
Who backs Wallet V
Wallet V is an incubation project of Virgo Group, a firm backed by a roster of crypto and venture investors: Draper Dragon, OKX Ventures, Vaulta Foundation, Cobo Ventures, Waterdrip Capital, and Sora Ventures. Those names signal that the project has both Asian and global capital behind it, though Wallet V itself remains a relatively small player in the crowded self-custody wallet space.
The benchmark release could change that — if the platform becomes a go-to testing ground for AI trading agents, the data itself becomes a draw. The next update, once it adds prediction markets and new model families, will tell whether the early results hold up or get revised.
