AI trading strategies: what a real-money live record shows

An AI trading strategy on real money returned +22.7% in 58 days in spring 2026. The number that matters sits right next to it: +12.9% over the benchmark in the same window. Finforge's Sophon-3 paper, Version 1.0, published in July 2026, reports a personal brokerage account trading Sophon Core, the firm's medium-risk agent, on the Nasdaq-100. It ran real capital from 27 April to 24 June 2026.

AI trading strategies: a real account, not a backtest

The account belongs to one of the paper's authors. It sits at Alpaca, an online broker, and the account number ends in 6414. It is the most verifiable evidence the paper holds, what the authors call Tier-4 evidence: real money, real fills, stated exactly. Every figure below comes from that account statement.

The benchmark that makes +22.7% honest

The account started at $1,772 on 27 April 2026 and ended at $2,174 on 24 June 2026. That is a portfolio return of +22.7%. Over the same window the Nasdaq-100 price index rose from a normalized $1,772 to $1,924, up +8.6%. Its total return, counting dividends, came in at about +8.7%. Against that total-return benchmark, the account's active return was +12.9%. A benchmark, not the buy button, decides what a return is worth.

What a 41-day record can and cannot prove

The window is about 41 trading days, roughly 0.16 years. The paper says plainly this is not statistically conclusive on its own. A short real-money account is real-fill evidence that the system runs as designed; it is not a verdict on edge. The account is small, so it says nothing about capacity. It runs the same Sophon Core configuration as the paper's simulated panel, with the differences that come from real fills and from a real brokerage crediting dividends.

Finforge runs four Sophon agents in public and publishes what they do, wins and losses. The agent pages trade founder capital in Alpaca paper accounts since April 21, 2026; this real-money account is separate and personal. You can check the live results, benchmarks and drawdowns for every Sophon agent on the agents page. Past performance is not a guide to future returns.

Questions people ask

Did an AI trading strategy earn real money?

A personal account trading Sophon Core on the Nasdaq-100 returned +22.7% from 27 April to 24 June 2026, or +12.9% over the total-return benchmark. It was real money in a real brokerage account, though the paper stresses the window is too short to be conclusive on its own.

Why compare an AI trading strategy to a benchmark at all?

Because a return without a benchmark is a guess. The Nasdaq-100 price index rose +8.6% over the same window, and its total return was about +8.7%. Measured against those, the account's +22.7% has an edge of +12.9% over the total-return index.

Does a short live record prove an AI trading strategy works?

No. About 41 trading days is not statistically conclusive on its own. The paper treats the account as evidence that the system operates as designed with real broker fills, not as a verdict on edge.

Published as a research paper. Coming to your phone.