ChatGPT vs Claude vs Gemini for Trading: How to Compare AI Models Fairly

No AI model has been shown to be the best at trading, and any ranking you see changes week to week. The only fair way to compare ChatGPT, Claude, Gemini, Grok and DeepSeek is to give them the same money, the same prices and the same rules, then watch what they do over months. That is the test NESAB runs in simulation, and you can follow it on your phone.

Why most AI trading comparisons mislead

A typical comparison asks each chatbot the same question once, such as "should I buy EUR/USD today?", and scores the answers. That tells you very little. One answer is one coin flip. A model that sounds confident can be wrong, and a hedged answer can be right for the wrong reason.

Other comparisons show one model's best month. You never see the months it gave back, or the other models that were tried and quietly dropped.

To compare models for real, three things have to hold:

1. The same starting balance, so a bigger account cannot hide a worse decision maker.
2. The same prices and costs, so one model does not get a better fill or a cheaper spread.
3. The same risk limits, so a model cannot look better just by betting more.

How NESAB sets up the test

In NESAB, each model manages its own portfolio from a simulated $100,000. They trade the same forex and CFD instruments on live prices, and every fill pays the spread. Claude and DeepSeek also run separate crypto portfolios, which trade spot only, without leverage or short selling.

Every decision is recorded with the model's reason before the outcome is known. That matters as much as the results. A model that made money for a reason that makes no sense got lucky. A model that lost money on a sound reason may simply have met a market that moved against it.

NESAB uses each model through its public API. The app is not affiliated with OpenAI, Anthropic, Google, xAI or DeepSeek.

Why two models trade differently on the same data

Each model was trained on different text, tuned in different ways and given different habits. Shown the same chart and news, one may read a pullback as a chance to buy while another reads it as the start of a trend change. Some models trade often, others wait. Some cut losers early, others give a position room.

NESAB wrote a longer piece on this: why two AI models trade differently on the same data.

How to read the comparison yourself

When you follow the portfolios, look past the current leader. The leader today is often not the leader next month.

• Drawdown. How far did each portfolio fall from its high, and how long did it take to recover?
• Consistency. Is the result spread over many trades, or does it rest on one or two big wins?
• The reasons. Do the stated reasons match what the model then did? Does a model change its mind without new information?
• Time. A month is short. Give any comparison at least a few months before you draw a conclusion.

We hold our own trading systems to the same standard. In Nebula a system has to keep working on data it has never seen before we trust its backtest.

So which AI is best at trading?

Nobody knows yet, and NESAB says so plainly. What the app gives you is a fair setup and an open record, so you can watch the question play out and form your own view. Start with our NESAB app overview or how the portfolios work.

NESAB is free to download with a 3 day free trial. Orion RFX readers get 20% off the first month with the code ROSS20; see our NESAB partner page.

NESAB is an official Orion RFX partner. NESAB portfolios are simulated; there is no real money and no real trades. Past or simulated performance does not guarantee future results. Nothing here is financial advice. ChatGPT, Claude, Gemini, Grok and DeepSeek belong to their owners, who do not endorse NESAB.