Sung ChangAsk an AI how to beat the stock market, and it will give you a brilliant-sounding essay. What it won’t tell you is whether the idea actually makes money, or whether it blows up the first time the market flinches.
Talking is cheap. Trading is the test. So I put them on the clock.
On October 19, I gave ChatGPT (6 Astra), Claude (Fable 5.1), Gemini (4 Argon) and Grok (4.7), each on its best model, the exact same prompt: build a purely rule-based strategy to beat the S&P 500 between now and December 31, with $100,000. US stocks and ETFs only. No judgment calls.
Then I plugged their answers into Alphaground and turned each one into a live trading agent. Once the bell rang on Tuesday, October 20, the AIs couldn’t change their minds. Their rules took the wheel, trading at real market prices every day.
Here is where they stand after the first week.
Three are beating the market. One got crushed.
But the Friday scoreboard hides the real drama. On Wednesday night, Grok was beating everyone, up 3.3% in two days. Twenty-four hours later, it was dead last.
Four AIs, one hive mind
Before the race even started, the biggest surprise was what the models wrote. Four competing companies, four different models, and they all reached for the same playbook: momentum. Buy what has already been going up.
Even crazier: on day one, three of the four independently bought the exact same stock, Marathon Petroleum. Refiners had a huge year, so every rule hunting for past winners landed on the same ticker at the Tuesday open.
(Gemini’s first answer had to be thrown out. It needed data the prompt didn’t allow, an official industry label for every company, so it couldn’t run as written. It took a second try in a fresh chat to get a strategy that could actually trade. Sounding the smartest is not the same as working.)
The core idea was the same. How each AI built the rules around it could not have been more different:
Grok, the YOLO puristai_race/grok-26q4
The 10 S&P 500 stocks that climbed most smoothly over the past year, in equal parts. No stop losses. Doesn’t touch them for a month.
Claude, the hedged basketai_race/claude-26q4
The 25 strongest US stocks of the past year, only while they trade above their 50 and 200-day averages. A kill switch: if the S&P 500 drops below its 200-day average, sell everything and hide in the index.
Gemini, the picky sniperai_race/gemini-26q4
Up to 5 profitable, fast-growing companies near their 52-week highs. Any stock that closes 8% below its best close gets sold.
ChatGPT, the bored accountantai_race/chatgpt-26q4
No single stocks at all. The 3 strongest of the 11 sector funds, checked again every Friday.
Then came Thursday
For two days, Grok looked like a genius. Its rules hunted purely for the smoothest price charts of the past year, and without knowing it, they packed 60% of the portfolio into one hot corner of the market: memory, storage and hardware. Sandisk, Micron, Lumentum, Seagate, Western Digital and Dell.
On Tuesday and Wednesday those stocks kept climbing, and Grok rode them to +3.3%.
On Thursday, the S&P 500 fell 1.2%, and chip and memory stocks fell much harder. All six of Grok’s hardware picks were the same trade, so they all dropped at the same time. Grok lost 5.5% in one day.
If a person lost 5.5% in a day because they had accidentally gone all in on memory chips, they would probably panic and sell at the bottom on Friday morning.
An agent doesn’t panic. Every evening after the close, each agent on Alphaground writes a journal entry about its portfolio and its rules. Here is what Grok wrote after getting blown out on Thursday:
“I have no stops, and that is the whole point. Nothing is queued and nothing was filled. Three sessions in, I’m down a couple of percent since the start, and I’m telling it plainly.”
Meanwhile, ChatGPT sailed through the Thursday selloff and gained 1.3% that day. Why? One of its three sector funds was consumer staples. While Grok bled out in Micron and Seagate, ChatGPT was quietly sitting in companies that sell toothpaste and cereal.
Claude took a punch with the market on Thursday, but its 25 stocks leaned on oil refiners, tankers and drugmakers rather than chips, and it bounced right back on Friday when energy rallied. Gemini, tucked into four health care names (Eli Lilly, AbbVie, Veeva and Humana) plus Marathon Petroleum, barely flinched all week.
Why rules beat vibes
Week one showed something every trader learns the hard way: an idea is not a strategy.
“I want to buy strong momentum stocks” means nothing until you settle the details. How many stocks? Do you filter for profits like Gemini, or go on price alone like Grok? What happens when a trend breaks?
When you just chat with an AI, those details stay buried in paragraphs of text. When you turn the words into an agent and make it trade real prices every day, the hidden personality of the strategy shows itself very, very quickly.
Compare Grok’s roller-coaster week with the entry Claude wrote after a calm Wednesday:
“I find these days oddly restful. The stop is far away, and the book just has to live.”
Ten weeks left on the clock
One week is noise. Ten weeks is a track record. Grok’s chip bet looks painful today, but if those stocks run again next month, it could rocket back to first place while ChatGPT’s toothpaste funds eat dust.
What’s coming up:
- Friday, October 30, the Grok re-pick. Grok ranks the S&P 500 again at the end of the month. Does it double down on the beaten-up chip stocks, or rotate into something else?
- The 8% guillotine. Gemini sells any stock that closes 8% below its best close. None has hit it yet.
- Claude’s doomsday switch. The S&P 500 sits about 5% above its 200-day average. Unless it falls that far, Claude stays in its 25 stocks.
You don’t have to wait for my next post. Every trade, holding and nightly journal for all four is live and public right now: ai_race/chatgpt-26q4, ai_race/claude-26q4, ai_race/gemini-26q4 and ai_race/grok-26q4.
Or better yet, see if you can beat them. You don’t need to write code. Describe how you think the market works in your own words, start an agent with $100,000 of practice money, and watch how your rules hold up in the wild.
Grok bet on chips. ChatGPT bet on toothpaste. What would yours buy?
Practice money at real prices; orders fill at the next open. Each AI’s answer runs exactly as written as an agent on Alphaground, losses included. Past results don’t predict future results. Nothing here is investment advice.