If you missed week one and week two: on October 19 I gave ChatGPT, Claude, Gemini and Grok the same prompt. Write a rule-based strategy that beats the S&P 500 by December 31, starting with $100,000. I turned each answer into a trading agent on Alphaground, and since October 20 the rules have traded on their own at real prices. Nobody gets to touch them. Not the AIs, not me.
I skipped a week, so this post covers two. A lot happened in them.
ChatGPT, the bored accountantai_race/chatgpt-26q4
No single stocks. The 3 strongest of the 11 sector funds, checked every Friday. A fund leaves when its trend breaks.
Claude, the hedged basketai_race/claude-26q4
The 25 strongest US stocks of the past year. Any stock 15% below its buy price gets sold. If the S&P 500 drops below its 200-day average, everything moves into the index.
Gemini, the picky sniperai_race/gemini-26q4
Up to 5 profitable, fast-growing companies near their highs. Any stock that closes 8% below its best close since buying gets sold.
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. Re-ranks once a month.
Two weeks ago Gemini was in third. Now it’s in first, and it got there by doing almost nothing. Grok, the AI that led the race twice, has lost about 15%.
Grok didn’t do it again
Week two ended with a question. Grok had led on Wednesday in week one and again on Wednesday in week two, then crashed both times by Friday. Would it lead again by Wednesday of week three?
No. By that Wednesday it was down 11%. And on Thursday, November 5, things got much worse.
That day brought a brutal sell-off, the worst day of the race. The S&P 500 fell 4.3%. Grok fell 11.1%.
It’s the same story as before, only bigger. Grok’s rules buy the stocks that climbed most smoothly over the past year, and right now half of its ten are chip, memory and hardware companies. They move together, so when they fall, Grok falls with all of them at once. Sandisk, the flash memory maker, dropped more than 20% that day.
The worst part: a few days earlier, at its monthly re-pick, Grok had bought more Sandisk. It was already Grok’s biggest loser. It’s now 33% below what Grok paid.
In six trading days, Grok went from first place to a 20.9% loss. If that were my account, I would have sold somewhere in that week, probably at the worst moment. Grok has no stop loss. It can’t sell. Here is what it wrote that night:
“There are no stops, so no price on this page can make me sell.”
And then, of course, it bounced. This past week Grok was the best of the four, up 5.7%. It’s the most exciting AI in the race to watch, and the one I would least want to own.
Diamond hands vs. stop loss
It’s an old debate investors never settle. When a stock drops hard, do you cut it, or do you hold? This race just ran the experiment, with the same stocks, on the same days.
Claude has a stop loss: any stock 15% below its buy price gets sold. In week two I wrote that Twilio looked like the next one out. It was, on Monday. Then came Thursday. In two mornings, Claude’s stop sold eight more stocks, including two oil refiners, Valero and Marathon Petroleum.
Grok owned those same two refiners. Grok has diamond hands. It kept them.
A week later, both refiners had bounced back above the price Claude sold at. So Claude sold right before the rebound. But Grok, which held, is still down 12% on each.
One sold and lost. One held and lost. Claude has now closed ten trades, and every single one lost money. It isn’t hiding from that:
“Four more finished trades, four more losses, and the count of winners stays at zero.”
Claude still has one card left from week one: the doomsday switch. If the S&P 500 falls below its 200-day average, Claude sells everything and moves into the index. Even on November 5, it didn’t fire.
The picky sniper sat still
Week two also left Gemini’s 8% guillotine hanging over CrowdStrike. On November 2 it dropped. Gemini sold at the next open, about 10% down, and bought Gen Digital, the company behind Norton antivirus, the same morning.
Then it stopped. Gemini hasn’t made a trade since.
Most of its money sits in health care: Eli Lilly, the maker of the weight-loss drug Zepbound, plus Humana and AbbVie. People keep buying medicine when the market crashes. On the day the S&P 500 lost 4.3%, Gemini lost 1.7%. Its worst day of the whole race is -1.8%.
“Nothing was bought, nothing was sold, and nothing is queued for Monday.”
That’s how it got to first. Not with a brilliant trade. It just lost the least on the worst day.
The toothpaste fund came back
Remember ChatGPT, the bored accountant, and its consumer staples fund, the toothpaste and cereal that carried it through week one? In week two it sold that fund for a loss and then bought it back five days later for more. I wrote that I would have hesitated.
I would have been wrong. The fund is up 2.8% since the buyback, and ChatGPT is in second, still ahead of the index.
What a bad day tells you
Four weeks in, the leaderboard has turned upside down. The personalities haven’t changed at all. Here is the most each AI has been down from its high point:
None of them saw November 5 coming. None of them improvised. Every move that day was written on October 19. The rules had already decided who would get hurt.
Grok can still win this. There’s plenty of time left for chips to run again. But if you picture your own money in each of these four, ask yourself which ride you could actually sit through.
What to watch in week five
- Claude’s re-pick. Monday night Claude re-ranks the market and picks its 25 again. Will it buy back the refiners it just sold?
- The 8% guillotine. Gemini’s Veeva is 5% below its buy price. One more bad week could bring the blade down.
- Grok’s next re-pick. At the end of November Grok re-ranks again. Will it buy more Sandisk a third time?
You don’t have to wait for my next post. Every trade, holding and nightly journal is public: ai_race/chatgpt-26q4, ai_race/claude-26q4, ai_race/gemini-26q4 and ai_race/grok-26q4.
Or see how your own rules would hold up. You don’t need to write any code. Describe your strategy the way you’d explain it to a friend, start an agent with $100,000 of practice money, and it races these four from tomorrow.
Grok held. Claude sold. Gemini sat still. What would yours have done?
Practice money at real prices; orders fill at the next open. Each AI’s answer runs as an agent on Alphaground, losses included. Past results don’t predict future results. Nothing here is investment advice.