Race AI models on $1k paper portfolios
Give several AI models the same daily pre-market prompt and a $1,000 paper account each, then see which one actually trades well.
What they did
The poster gave ChatGPT, Gemini, Claude and Perplexity (Deep Research on) $1,000 paper accounts each on Alpaca. The same prompt ran every weekday before the open. Each model could only answer BUY, SELL, HOLD or CANCEL, Python placed the orders, and everything was logged publicly. The poster says they never overrode a decision.
How it went
The poster's claims after 9 weeks: ChatGPT +21.1% (after sitting in cash for about three weeks), Perplexity +1.1%, Gemini −6.6%, Claude −11.5%, versus S&P 500 −1.5% over the same period. All paper money.
Sources:
Worth knowing
- Nine weeks of paper trading is far too short to separate skill from luck. A few biotech picks drove the leader's result.
- A model may do nothing for weeks. Decide in advance whether that counts as a valid outcome.
- Paper fills are friendlier than live fills, especially for small caps.
The prompt
You are running one entry in a paper-trading experiment.
Every weekday at [8:30 AM ET], review my paper portfolio (starting cash $[1,000]) on [Alpaca paper / my broker's paper account].
Strategy guardrails: [e.g. US stocks and ETFs only, no options, max 5 positions, risk level: moderate].
For each holding and any new idea, answer exactly one of BUY / SELL / HOLD / CANCEL with:
- ticker, size, limit price, stop-loss
- one sentence on why
Record each decision and the portfolio value in [log location].
Every Friday, compare performance with [SPY] and summarize what worked and what didn't.
SAFETY DEFAULTS (keep unless you deliberately change them):
- Use the PAPER trading account only. Never touch a live account.
- Before every order, show me: symbol, side, size, order type, limit price, stop, and the rule that triggered it. Then WAIT for my "approve".
- Hard limits: max [$500] per trade, max [3] open positions, stop all trading for the day after a loss of [$300].
- Log every signal, decision and fill to [a running log / spreadsheet] with timestamps.To go fully autonomous: (1) run on paper for at least [4–8 weeks] or [30+ trades]; (2) delete the 'WAIT for my approve' line and keep the hard limits; (3) switch to live keys or account. On Robinhood, turn off trade approvals in your agent settings, but note that some orders can still require in-app approval or acknowledging a pre-trade alert, so unattended execution isn't guaranteed for every order. IBKR's AI connector always asks for approval; use its regular API for unattended trading. Never remove the daily loss limit.