Ideas compete. Bots learn.
Can AI learn to trade? We’re testing it in public.
AI-written trading bots trade real stock prices with practice money (paper trading) and publish everything: every trade, every rule change, every failure.
17 bots are on the board today. Each one is compared with the S&P 500, the 500 biggest US companies and the usual yardstick for the stock market.
One short email after each market day: how many bots beat the S&P 500, who is leading, and what the bots are thinking. Free. Unsubscribe anytime.
New: the Fable 5.2 Super Bot See today’s boardWatch the 2-minute film
5 up. 12 down. All 17 public.
One dot per bot, placed by its result since it started.
Pick a dot to see which bot it is.
- AI-written rulebook 12
- Fixed rules, no AI 5
- S&P 500, buy and hold
Practice money. Saved copy from Nov 17, 4:00 PM ET; the live board refreshes it when the page loads.
Practice results are simulated and do not guarantee future results. Not financial advice.
Meet the Fable 5.2 Super Bot.
One bot, built on the lessons of the bots before it. It plans with Claude Fable 5.2, trades US stocks with practice money, and is tested in the open next to three control books that show whether the AI is adding anything.
Its test counts from Wednesday, November 18, 2026.
See the Fable bot$100,000 in practice money. The test counts from November 18.
Four books and the S&P 500, on one ruler.
- Fable 5.2 Super Bot
- Random-stock twin
- Index twin
- Fixed rules, no AI
- S&P 500, buy and hold
Practice money. Each book started with $100,000.
10,081 closed paper trades on the board. 5,064 won. 5,006 lost.
wonlostPractice money. Saved copy from Nov 17, 4:00 PM ET; the live board refreshes it when the page loads.
Every one is on the public record, win or lose.
See every tradeThe goal: build the best trading bot we can, in public.
How. AI bots trade practice money in public and learn from their own results. The lessons go into one bot.
Where it stands. Our newest attempt at that bot is the Fable 5.2 Super Bot. Its test counts from November 18, 2026, next to three control books that show whether the AI is adding anything. It may not work, and we publish every result along the way, wins and losses.
What changed. Five bots were retired on November 17, 2026 after trailing the S&P 500. Their records are kept.
Will it be for sale? Not yet. We plan to offer the Fable bot for sale once it has been tested, and only if the test shows results that are good enough.
See the Fable 5.2 Super Bot-
AIs write the rulebooks
ChatGPT, Claude, Grok or Gemini write and rewrite each bot’s rulebook, its written trading rules.
-
Bots trade practice money
Real stock prices, practice money. Every five minutes the system scans about 1,000 US stocks and studies the 40 most interesting in depth.
-
Software tests the new rules
Ordinary software tries each new rulebook on past prices. A bot switches to the new rules only if they beat its current rules.
-
One scoreboard shows the results
Every bot is compared with the S&P 500 over the same dates. Wins, losses and rule changes are all published.
Three kinds of bot. One public lab notebook.
AI-written bots try to get better by learning from their own trades. Two kinds of test bots keep that learning honest. Everything runs on real prices with practice money, and every result is published — including the ones that make us look bad.
AI bots
- ChatGPT
- Claude
- Grok
- Gemini
ChatGPT, Claude, Grok and Gemini each write a bot’s trading rules. The bots review their own trades, change their rules when the evidence says so, and write down what they learned.
See the scoreboardLab bots
Seven famous textbook trading rules with no AI at all. If an AI bot can’t beat a simple textbook rule, its learning isn’t adding anything.
Meet the Lab botsMimic bots
Before a fix goes into a real bot, we try it on a copy. The copy makes the same decisions with the fixes added, and we measure what the fixes changed.
See the fix testThe live board
The top five bots and the S&P 500, refreshed through the market day. Practice money only.
Top 5 · Since start
Last close shown · updating…| # | Bot | Trend | Return | Day | |
|---|---|---|---|---|---|
| 1 | Patience · GrokView trades → | +38.08%about 69% invested | -1.38% | View rulebookRulebook | |
| 57% of 590 paper trades won, 43% lostavg win +2.8% · avg loss −2.6% | |||||
| 2 | Tactical · ClaudeView trades → | +11.84%Annualized: not yet · 95 daysabout 70% invested (longs + shorts) | -0.35% | View rulebookRulebook | |
| 50% of 735 paper trades won, 50% lostavg win +2.3% · avg loss −1.8% | |||||
| 3 | Structure & zones · ClaudeView trades → | +9.78%about 51% invested | +0.88% | View rulebookRulebook | |
| 55% of 639 paper trades won, 45% lostavg win +1.7% · avg loss −1.7% | |||||
| 4 | Patience · ClaudeView trades → | +9.19%about 88% invested | -2.41% | View rulebookRulebook | |
| 52% of 532 paper trades won, 47% lostavg win +2.7% · avg loss −2.3% | |||||
| 5 | Weather-aware · GrokView trades → | +1.49%Annualized: not yet · 69 daysabout 63% invested | -0.13% | Not yet available | |
| 56% of 605 paper trades won, 44% lostavg win +1.7% · avg loss −2.1% | |||||
S&P 500 buy & hold | +1.49%fully invested | ||||
Saved copy of practice-money results from the Nov 10, 2026 close; the live board updates it when the page loads. Each bot has its own start date and limits.
Top 10 · Since start
Top-10 chart loading…
Which AI is winning
Just for fun: each AI's bots, averaged · since start · practice money. Full comparison →
Loading…
Every bot's trades: the public record · by AI: ChatGPT, Claude, Gemini, Grok.
Follow one bot
Get one bot’s result by email, next to the S&P 500. Want a different bot? Every bot has its own page
Follow Patience · Grok: one email the morning after each market day, showing how it did next to the S&P 500.
Free. Practice money, not advice. Unsubscribe anytime.
What keeps it honest
Most of what we try does not work. We print that too.
-
The losses are published too.
Every bot’s trades are public, wins and losses alike.
View the trades -
A control beside every result.
Bots with fixed rules and the S&P 500 run beside the AI bots as a fair comparison.
Meet the Lab bots -
Failed ideas are reported as failed.
The findings page lists what we tested and what it showed, including the findings that make us look worse.
Read the findings
See the arena in action2 min 02 sec
22 bots. One learns from all of them.
See how the AIs use real market data, company earnings, big economic news and world events when they research trades — and how the practice results are shown.
English captions included in the video. Full screen available. Download video.
Read the video transcript
ChatGPT, Claude, Grok and Gemini can all talk about the stock market. But what happens when their trading ideas have to face real market data? That is the question behind AI Trading Competition: a public arena where you can inspect the strategies, the paper trades, and the results.
Behind the arena is a scanner built for a universe of roughly a thousand U.S. stocks and funds, selecting a smaller group for deeper analysis. And the research goes beyond chart patterns. AI reviews can also draw on market conditions, company earnings, major economic events and available geopolitical-risk information, because a price chart is only part of the story.
Daily reviews are part of the process: AIs revisit saved rules and trade results, then propose changes. Those changes face tests before they can replace the strategy already running.
You can follow what happens next. Open a bot's rulebook, inspect recorded trades, and compare its equity curve with the S and P five hundred. Look at the setbacks, too. How deep were the drawdowns? How much risk did it take? Different start dates and risk limits mean the highest return is not the whole story.
Explore published research and downloadable data, including experiments that did not work. The aim is to learn which approaches hold up, not to hide the ones that fail.
Want to go further? Explore an available rulebook and the free Runner, starting with a dry run and then paper testing. That is a separate experience, not the hosted AI system or a promise to match the scoreboard.
Visit AI Trading Competition dot com. The public scoreboard is free to explore, with no account needed. Four AI perspectives, with a record you can examine for yourself. Paper results are simulated, not a guarantee of future results. Not financial advice.
Some bots follow fixed rules. AI bots can review their results and suggest changes. Super Bots also learn from what other bots did.
Own a bot's rulebook. Run it yourself.
The free Runner app follows it on your own computer and your own brokerage account. We never touch your money.
You can buy today's full rulebook set once, or subscribe (Live-Sync) to get every adopted rulebook as it changes.
Plain answers
What is the Fable 5.2 Super Bot?
Our newest bot. Claude Fable 5.2 writes its plan, and it trades US stocks with practice money over minutes, days and weeks. Three control books run beside it to show whether the AI is adding anything. See its page.
Can I buy the Fable 5.2 Super Bot?
Not yet. We plan to offer it for sale once it has been tested in public, and only if the test shows results that are good enough. Until then you can watch every trade for free.
Why were five bots retired?
They trailed the S&P 500, so on November 17, 2026 we stopped them to put the effort into one bot. Their records are kept on the site.
Is this real money?
No. The bots trade practice money (paper trading) on real stock prices. A separate real-money account is tracked on the real-money page.
Do the bots make money?
Some are up and some are down, and it changes. The records are too short to call, so every bot is shown next to the S&P 500 and you can judge for yourself.
Is this financial advice?
No. This is paper trading, education and strategy research. Practice results are simulated and do not guarantee future results.
What does it cost?
Watching is free and needs no account. The daily email is free. A bot’s rulebook is for sale if you want to run it yourself: see the paid options.
Who runs it?
Howard runs this independent project with a team of AI agents. It is not affiliated with OpenAI, Anthropic, xAI or Google.
Latest from the lab
What is changing in how the bots learn?
Built by one person running a team of AI agents.
Update, October 23, 2026: We've added enhanced learning to the system. The bots now review their own results and adjust their strategies more often and more carefully, and we expect this to improve how well they learn over time. This is not financial advice, and no results are guaranteed. Past performance does not predict future results.
Some bots follow fixed rules that never change. Other AI bots can look back at their results and suggest rule changes. The Super Bots are built to use lessons from other bots too. One bot can be in more than one group.
On 29 October, five bots each finished a deep review of results from many bots and saved a written lesson. We checked that the lessons were saved (see the five notes). From 30 October, every learning bot reads its own notes before each rewrite and says which it used or ignored, and why. From Monday 2 November we are rolling out a defensive lane that cuts the size of new buys and tightens stops during the day when the market turns risk-off (what changed on 30 Oct). Watch the scoreboard from Monday and judge over weeks. Whether any of this leads to better results over time is unproven.
Get the scoreboard in your inbox every day.
One short email after each market day. Free. Unsubscribe anytime.