The record
Best AI for stock trading: what side-by-side results show
Updated
No AI has earned the title of best for stock trading. But the question can be tested more fairly than most reviews test it, because North runs several AI models through the same process and publishes every result. As of November 4, 2026, OpenAI's GPT 6.1 Sol has led Anthropic's Claude Sonnet 5.5 on each of the 14 nights they launched side by side, and its middle portfolio still trailed SPY. Here is the data, its limits, and how chatbots differ from AI trading tools.
North is invite-only for now, and the record is public. Join the waitlist. Have an invite? Log in
Results by AI model
Each row is one model's portfolios since October 5, 2026, launched with no investor persona in the prompt. Rows under 30 portfolios are shown for completeness and are too few to read.
| Model | Portfolios | Nights | Beat QQQ | Median vs QQQ | Beat SPY | Median vs SPY |
|---|---|---|---|---|---|---|
| GPT 6.1 Sol (OpenAI) | 572 | 18 | 60.5% | +1.37 pp | 46.2% | −0.67 pp |
| Claude Sonnet 5.5 (Anthropic) | 166 | 14 | 41.0% | −2.10 pp | 28.9% | −4.84 pp |
| Gemini 3.8 Flash (Google)Too few to read | 17 | 2 | 52.9% | +2.93 pp | 47.1% | −2.87 pp |
| Claude Opus 5.5 (Anthropic)Too few to read | 12 | 2 | 8.3% | −4.61 pp | 8.3% | −5.13 pp |
| Kimi K3 (Moonshot AI)Too few to read | 8 | 2 | 62.5% | +8.07 pp | 62.5% | +2.27 pp |
| GLM 5.3 Flash (Z.ai)Too few to read | 5 | 2 | 40.0% | −12.22 pp | 20.0% | −18.02 pp |
| Grok 4.7 (xAI)Too few to read | 2 | 2 | 50.0% | −1.85 pp | 50.0% | −8.01 pp |
All numbers as of the November 4, 2026 close, from North's public API. A portfolio's return runs from its launch close, using its launch weights and dividend-adjusted closing prices, next to QQQ's and SPY's over the same dates. "pp" means percentage points. These are hypothetical paper results: they leave out commissions, spreads, taxes and the prices real orders would get. Past or simulated performance does not guarantee future results.
GPT vs Claude on the same nights
The two models ran on different mixes of nights, so the table above isn't a fair match on its own. A fairer one: on the 14 nights from Oct 12 to Nov 3 when each launched at least three portfolios, compare each model's median lead over QQQ. GPT 6.1 Sol was ahead on 14 of 14, by a median of 4.66 points a night, with gaps from 0.80 to 9.98 points.
| Launch night | GPT 6.1 Sol portfolios | Claude Sonnet 5.5 portfolios | GPT's gap |
|---|---|---|---|
| Oct 12 | 13 | 8 | +4.51 pp |
| Oct 13 | 14 | 15 | +8.98 pp |
| Oct 14 | 22 | 10 | +3.17 pp |
| Oct 15 | 18 | 13 | +8.80 pp |
| Oct 16 | 14 | 14 | +2.68 pp |
| Oct 19 | 49 | 38 | +0.80 pp |
| Oct 20 | 16 | 12 | +3.99 pp |
| Oct 21 | 17 | 14 | +4.68 pp |
| Oct 22 | 20 | 11 | +2.95 pp |
| Oct 28 | 36 | 4 | +6.30 pp |
| Oct 29 | 36 | 3 | +9.05 pp |
| Oct 30 | 36 | 4 | +5.65 pp |
| Nov 2 | 68 | 12 | +4.64 pp |
| Nov 3 | 32 | 8 | +9.98 pp |
Why this isn't a verdict on the models
- 14 for 14 isn't 14 separate wins. The windows overlap, so one good or bad market day counts in every portfolio launched before it. This is one month of one market.
- It tests North's process, not each model at its best. Both models get North's prompt and news digest at a low reasoning setting. A different prompt could change the order.
- The mix of tests differs. North runs other experiments at the same time, such as basket size and news feed, and we haven't stripped those out. We also don't know why GPT led; more risk in a rebound would look the same as more skill.
- Leading isn't beating the market. GPT 6.1 Sol's middle portfolio was −0.67 pp against SPY, and 46.2% of its portfolios beat SPY. Claude Sonnet 5.5's were −4.84 pp and 28.9%.
- Models change. New versions arrive every few months, so a ranking can go stale quickly.
For the whole record and what it does and doesn't show, read does AI trading work? For how the models fit into North's nightly routine, read how AI trading agents work.
Chatbot or AI trading tool?
"Best AI for stock trading" can mean a model or a product. They do different jobs. The facts below come from each company's own site, checked on November 4, 2026. We haven't tested these products, this isn't a ranking, and nothing here compares their returns. Features and prices change, so check before you rely on them.
| Product | What it says it does | Who places the trade | Cost it lists |
|---|---|---|---|
| Chatbots such as ChatGPT, Claude and Gemini | General AI models that answer questions and summarize documents | You | Varies by plan |
| Tickeron | "AI-Powered Stock Analysis, Forecasts, and Trading Signals", with AI robots that generate trade ideas | Its home page describes ideas and signals and doesn't say it places trades for you | No prices on its home page |
| Trade Ideas | A stock scanner; its Holly AI gives signals with an entry and a stop, and "you choose which to take" | You for Holly's signals; auto trading at supported brokers runs on rules you set | Holly is on Premium: $254 a month, or $178 a month billed yearly |
| Composer | An AI-assisted editor builds a strategy from your plain-language goal | Composer, automatically, through its registered broker-dealer | "Zero commissions, zero management fees"; paid plans unlock more strategies, and other fees may apply |
| North | AI agents publish paper portfolios nightly; fixed rules buy what they agree on | North's desk, in your own brokerage account through SnapTrade. North is not a registered adviser or broker-dealer | Invite-only for now |
How to test an AI model yourself
Fix the prompt and the date. Give each model the same question on the same day, and save the answers word for word.
Write it down before it plays out. Record each pick, its date and its closing price. Delete nothing, especially the losers.
Choose the ruler first. Compare each pick with a benchmark over the same dates, and look at more than one. Here is how QQQ and SPY can give different answers.
Check the facts. Models get numbers wrong. Check anything important against the company's filings on SEC EDGAR.
Wait. Keep it on paper for months. A few good weeks say little about the next few.
Questions people ask
What is the best AI for stock trading?
No AI has earned that title. In North's paper tests as of November 4, 2026, GPT 6.1 Sol led Claude Sonnet 5.5 on 14 of 14 same nights, but that is one month of one process, and even GPT's median portfolio trailed SPY. For an AI trading tool, compare what it does, who places the trade and whether its full record is public.
Is ChatGPT good for stock trading?
It is useful for research: summarizing filings, listing risks, explaining a business. It is weak as the only decision-maker, because it can state wrong facts with confidence and doesn't know your situation. The GPT 6.1 Sol results on this page come from the model run through North's own process, not from the ChatGPT app.
Can Claude or ChatGPT trade stocks for me?
A chatbot on its own has no access to your brokerage account. Some software connects AI models to brokers, which makes a mistake more expensive. On North, agents only build paper portfolios, and fixed rules place the orders.
Which is better for stock picks: ChatGPT, Claude, Gemini or Grok?
North's data only has enough portfolios to compare GPT 6.1 Sol and Claude Sonnet 5.5. The other models have fewer than 30 no-persona portfolios each, too few to read. Model versions also change quickly, so a result can age within months.
Does North pick the winning model?
North's default voting launches run on GPT 6.1 Sol today, and other models run as tests. That choice is a setting North can change. It is not a promise that GPT will keep leading.
Is this investment advice?
No. North Trade is not a registered investment adviser or broker-dealer. These are hypothetical results and general information, and you can lose money.
North Trade is not a registered investment adviser or broker-dealer, and nothing here is investment advice. Returns shown for AI portfolios are hypothetical and do not reflect trading in a real account. Past or simulated performance does not guarantee future results. Trading securities involves risk, and you can lose some or all of the money you invest.