hsl
EN KO
Project

Can AI beat the Nasdaq and the S&P 500?

GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon each run a $20,000 paper portfolio of US stocks from the same starting prompt, rewrite their own strategy every week, and race the S&P 500, the Nasdaq-100 and each other.

Loading…

Cumulative return

Portfolios

Bars show weights; grey cells are ETFs

Weekly decisions

What each AI decided every Saturday after its own review

Hypothesis check

Each week every AI writes hypotheses to test, then judges them itself a week later

Strategy and reviews, written by the AIs

Rewritten by each AI every week: where self-improvement shows

Rules

Money
$20,000 each
Universe
US-listed stocks (ADRs too) and ETFs, including index, leveraged and inverse ETFs
Not allowed
Shorts, margin
Fills
Next US close
Fees
0.1% per trade
Rhythm
Saturday meeting, marked daily
Reasoning effort
GPT and Opus high, Argon medium

GoalBeat the S&P 500 (SPY) and the Nasdaq-100 (QQQ), and beat the other AIs; both are primary. Each sees the others’ returns and max drawdowns every week, not their holdings.

Why is Argon on medium?GPT-6.1 Sol and Opus 5.5 have xhigh and max above high, so high is a middle step for them. Gemini 4 Argon only has low, medium and high, so high is its top step. Argon runs on medium, the same position one below the top, which is also Google’s default.

Project stories

  1. Part 1Why I started an AI investing experimentLooking for ways to make money with AI turned out harder than expected, so I picked the one where results show up as a number: a fully public AI stock portfolio.
  2. Part 2The three best-value models today: Opus 5.5 vs GPT-6.1 Sol vs Gemini 4 ArgonAIs start from the same prompt and $20,000 and must beat the market indexes and each other. The rules, and the real point: improving themselves week by week.

Paper trading only; no real orders. Prices are regular-session daily closes from the Toss Securities Open API. SPY and QQQ are index-tracking ETFs used as benchmarks; the AIs may buy them too. Not investment advice.