Can ChatGPT Trade Crypto? What Happened When AI Models Traded Real Money
A chatbot can place trades if you connect it to an exchange, but there is no evidence it trades well. In nof1's Alpha Arena, six leading AI models each traded $10,000 of real crypto for about two weeks in late 2025, and four of them lost money; GPT-5 lost about 60%. In our own pre-registered test, a frontier model chose which of 499 real trade setups to take or skip, and its choices did no better than skipping at random. AI language models are good at explaining, summarizing and following a process. Picking trades from a chart is not one of their skills, so Botsfolio's AI runs strategies tested on years of data and uses the language model to explain them.
"Can ChatGPT trade crypto for me?" now has 80-plus variations in search suggestions, and YouTube is full of people handing a chatbot their account. The answer depends on what you mean by "trade". A language model can certainly send orders. Whether its orders make money is a different question, and it has been tested with real money.
When six AI models traded real money
In October 2025, the research lab nof1 gave six leading AI models $10,000 each to trade crypto perpetual futures on their own, with every trade public. The contest, called Alpha Arena, ran from October 18 to November 3:
Two Chinese models finished ahead: Qwen3 Max returned about 22% and DeepSeek about 5%. The other four lost between roughly 30% and 60%. Commentators drew the obvious conclusion: two weeks, leverage and one market regime say more about luck than skill, and most of the field lost money.
Our own test: can a frontier model pick better trades?
Alpha Arena asked models to trade from scratch. We asked a narrower question that should be easier: given a real trade setup, can a frontier AI model tell good ones from bad ones?
We gave a leading language model 499 real setups from strategies we publish, from July 1 to October 5, 2026. For each one it saw an anonymized chart and market context and answered take or skip. We wrote down the test and the pass mark before running it. To pass, its choices had to beat at least 95% of random skip sets of the same size.
| 499 real setups, July 1 to October 5, 2026 | Result |
|---|---|
| Trades the model chose to take | 276, averaging +0.149R |
| Trades it chose to skip | 222, averaging +0.001R |
| Total result, its picks vs taking every trade | +41.0R vs +41.2R |
| Random skips that did as well or better | 128 of 1,000 tries |
| Link between its confidence and the result | None (rank correlation +0.03) |
| On 150 older trades | Reversed: taken -0.148R, skipped +0.138R |
On the main set, the trades it took did better on average than the ones it skipped. But the total result was the same as taking every trade, 128 of 1,000 random skip sets did as well or better, and its confidence had no link with how trades turned out. On an older set of 150 trades, its choices were worse than its skips. Taken together, the gap is noise. It failed the test.
Why a chatbot can't trade from a chart
- It has no tested edge. A language model writes plausible text. A confident "this setup looks strong" is not a measured advantage.
- It sounds sure when it's wrong. People who connected chatbots to accounts have found them trading on stale or invented prices. A fluent explanation makes a bad trade look reasoned.
- It can't be tested the usual way. You can't backtest a chatbot's judgment over years of data the way you can a fixed strategy, so you have no record before you risk money.
- Markets punish confidence. Two weeks of leverage can turn a confident model into a 60% loss, as Alpha Arena showed.
None of this means AI is useless in trading. It means the job matters.
What AI is actually good at in trading
- Running a tested process without emotion, the same way on the hundredth trade as on the first.
- Explaining what a strategy did and why, in plain language.
- Watching many coins and many stops at once, around the clock.
- Keeping the record of every trade, honestly.
That is how Botsfolio uses it. The AI agent runs strategies tested on years of data, judged against simply holding the same coins, and the language model explains what it did. It never picks a trade from a hunch. You can check the result on the live watch page:
All 7 portfolios are ahead of holding the same coins, and every one fell less than holding did at its worst.
| Portfolio | Return | Biggest drop |
|---|---|---|
| BTC & ETH Cycle Trader BTC · ETH | +24.1% hold -13.2% | 8.7% hold 46.5% |
| BTC & ETH Trend Rider BTC · ETH | +3.9% hold -13.2% | 6.1% hold 46.5% |
| BTC & ETH Steady Trader BTC · ETH | +2.6% hold -13.2% | 3.2% hold 46.5% |
| Bitcoin Cycle Trader BTC | +3.4% hold -7.5% | 4.8% hold 39.6% |
| Crypto Cycle Trader BTC · ETH · ZEC · XRP · SOL | +29.1% hold +11.9% | 12.4% hold 45.5% |
| Altcoin Trend Rider ZEC · XRP · SOL · XLM · NEAR | +34.8% hold +18.1% | 13.4% hold 50.2% |
| Memecoin Trend Rider DOGE · FARTCOIN · PENGU · PEPE · BONK | -6.3% hold -36.8% | 10.5% hold 63.2% |
YouTube tutorials offering a free ChatGPT or Claude crypto bot are a known scam. TRM Labs traced one campaign that tricked 224 people into deploying contracts that drained their own wallets. More in are AI trading bots legit.
FAQ
Watch Botsfolio AI trade live
Botsfolio AI paper trades on the live market, and every entry, stop and exit is public.
- BTC & ETH Cycle Trader+24.1% vs -13.2% buy & hold in 2026 · biggest drop 8.7% vs 46.5%
- BTC & ETH Trend Rider+3.9% vs -13.2% buy & hold in 2026 · biggest drop 6.1% vs 46.5%
- BTC & ETH Steady Trader+2.6% vs -13.2% buy & hold in 2026 · biggest drop 3.2% vs 46.5%
- Bitcoin Cycle Trader+3.4% vs -7.5% buy & hold in 2026 · biggest drop 4.8% vs 39.6%
- Crypto Cycle Trader+29.1% vs +11.9% buy & hold in 2026 · biggest drop 12.4% vs 45.5%
- Altcoin Trend Rider+34.8% vs +18.1% buy & hold in 2026 · biggest drop 13.4% vs 50.2%
- Memecoin Trend Rider-6.3% vs -36.8% buy & hold in 2026 · biggest drop 10.5% vs 63.2%
How to read these numbers: the buy and hold comparison, biggest drop, and how the portfolios are tested.
See everything Botsfolio AI trades live
Paper trading on real prices. Not investment advice.
Educational analysis, not financial advice. Alpha Arena figures are as reported by press coverage. Our test results are research on past trades. Trading crypto can lose money. Past performance does not predict future results.