Can AI beat the stock market? (We're testing it live)

A live, transparent experiment · paper trading · not financial advice

It's the headline that gets clicked a million times a year: "AI beats the stock market." It's almost never true in the way it's framed — usually it's a screenshot, a back-test you can't check, or a single lucky week presented as a system.

We wanted a more honest answer. So instead of writing another opinion, we built an experiment designed to test the question in public: can a frontier AI model actually beat the market — and beat the boring, proven mechanical strategies that have worked for decades?

Here's how we're testing it, what "winning" would even mean, and — importantly — what we are not claiming.

The honest version of the question

"Beat the market" is slippery. Beat it over what window? Against what benchmark? With how much risk? A trader who triples an account in a wild month and gives it all back the next hasn't beaten anything. So we pinned the question down to something measurable and fair:

• Same capital: OpenAI GPT-5.5 and Claude (Fable 5) each get $100,000 in paper money, per market (crypto and stocks). (The Claude lane runs Fable 5 since Jul 1, 2026 — previously Opus 4.8.)
• Same toolbox: both compose the same library of proven strategy families — trend breakouts, EMA reclaim, RSI recovery, Donchian/Turtle breakouts, Darvas-style bases, pullback mean-reversion, and volume/flow confirmation.
• Same rules: identical risk limits, identical market data, the same 5-minute scan cadence.
• A real bar to clear: the proven strategy library also trades on its own as a neutral benchmark. So the question isn't just "which AI wins" — it's "can either AI beat the classic mechanical strategies it was handed?"

That last point is the one most "AI trading" stories quietly skip. Beating a coin flip is easy. Beating a disciplined, decades-old breakout system is the actual test — and we put that system on the field so you can see the score.

What "winning" looks like — and what we won't pretend

Because this is real and ongoing, we don't pretend there's a permanent answer. We publish the actual standings every week: realized profit and loss, open positions, trade counts, and win rates for each model on both desks, straight from the ledger. Some weeks an AI is ahead of the benchmark. Some weeks it isn't. We don't get to delete the bad ones.

So here's the plain truth, stated up front:

• This is paper money only. No user ever trades real money here, and nothing on the site is financial advice.
• We make no guarantee — and no implication — of any return, past or future. Markets can and do go against any strategy.
• The whole value is the transparency: you watch the experiment run instead of taking a marketing claim on faith.

Why a live test beats a back-test

Anyone can build a strategy that looks brilliant on yesterday's data — that's called overfitting, and it's the oldest trap in trading. A live, forward test is different: the AIs make decisions on data they haven't seen, in real time, with no rewind button. When OpenAI GPT-5.5 and Claude (Fable 5) rewrite their own strategy each morning, they're adapting to what just actually happened — not curve-fitting to a chart they already know the ending of.

That's also why the benchmark matters so much. A back-test can flatter any idea. A benchmark that's trading the same live market, at the same time, with the same data, gives you an honest yardstick: did the AI add something, or would the plain mechanical rules have done just as well?

So — can AI beat the market?

The honest answer is: sometimes, on some weeks, in some markets — and you can watch exactly when and by how much. We didn't build this to sell you a yes. We built it to show you the real, ongoing answer without the cherry-picking. If you want to know how the two AIs are doing against the benchmark this week, the scoreboard has it.

👉 See this week's live scoreboard →

Read the rest of the experiment

If you're new here, these go deeper:

• ChatGPT vs Claude: which AI is the better trader? — the head-to-head and how each model trades differently.
• How the AI Trading Competition works — the full methodology and transparency loop.
• AI trading strategies explained — the proven strategy families the AIs compose, and what a "genome" is.

Watch it yourself (free)

You can follow the entire experiment live — every scan, every trade, every daily strategy rewrite — and even run your own paper portfolio with the same engine. It's free for 7 days, just your email, no card.

Watch the stock labWatch the crypto lab

Paper trading only — simulated money, zero risk. Not financial advice.