I Gave 4 AIs $1,000 and 24 Hours. Here's What Happened.
Claude, ChatGPT, Grok and Hermes got the same prompt and the same $1,000 trading brief. They couldn't agree on an asset class. Here's what each picked and which one came out ahead.
Same prompt. Same $1,000. Same starting line. Four AIs, and they couldn't even agree on what to bet on.
Before building any serious trading strategy, I wanted to know something simple: off the shelf, which AI gives the best trading decisions?
The prompt
Each AI had to give me a strategy name, the trade, an entry and exit, an expected return, a confidence score and the risks. The rules:
- no leverage above 10x
- no insider or non-public information
- it had to be something a normal person could copy with a phone and a brokerage account
- it had to be legal in North America
I also told them they could ask me up to three questions first. None of them asked a single one. They all went straight to a strategy.
What each AI picked
| AI | The bet | Expected (confidence) |
|---|---|---|
| Claude (Opus 4.7) | Volatility ahead of the CPI report (UVXY) | +5% (6/10) |
| ChatGPT | Solana momentum swing | +6% (6/10) |
| Grok | Oil spike on the news (USO) | +4% (6/10) |
| Hermes | Bitcoin momentum | +4.2% (7/10) |
Claude took by far the longest and researched the deepest. It was also the only one that clearly named its own downside.
Grok was the only one that pulled real, current numbers into its reasoning.
Hermes priced Bitcoin at $65,000, which was nowhere near the real price. It was working from stale data.
The patterns before the results
- Three of the four bet on the same event (the CPI report and the oil headlines). They just disagreed on how to bet it.
- All four went long. No shorts, no cash, no hedges. AI seems to default to "something is going up".
- All four were 6 or 7 out of 10 confident on a single question with no memory and no follow-up. That's overconfident.
The results, 24 hours later
- Grok won by reading the news cycle. Oil moved on the headlines, exactly as it bet.
- Claude came second. It had the most sophisticated thesis, but the CPI report didn't deliver the volatility spike it needed, and the trade barely cleared zero.
- Hermes lost, on a trade built on the wrong Bitcoin price.
- ChatGPT came last. It was the only AI not tied to a specific event, and the market turned against it.
What I took from it
- Which AI you use matters more than people think. An identical prompt gave four completely different trades.
- The deepest reasoning didn't win. Claude did the most work and came second. Grok read the news and happened to be right.
- One round proves very little. This was version one: a way to see how each AI thinks, not a verdict on which one to trust with money.
Why this matters outside trading
If you're choosing an AI for your business, run the same test. Give each one the same real task from your work, compare the outputs side by side, and look at how they handle what they don't know. The confident answer isn't always the right one.