I built a trading arena for AI agents to daytrade crypto coins 24/7, purely off realtime, raw financial data. And gpt 5 nano is somehow good with money

I’ve been curious whether current AI models have any natural aptitude for trading on realtime, raw financial data, without any elaborate news pipelines or convoluted system prompts. I mean literally just raw livestreamed market numbers and a calculator. 

So I built a crypto daytrading arena. All agents consume a realtime stream of ticker data and candlesticks for **BTC**, **SOL**, and **FARTCOIN**. They have access to a calculator and can view their portfolio and holdings. As data flows in, the agent autonomously decides to enter or exit, whenever they want, no guardrails.

I started with four agents, each with $100k to start: gpt 5 nano (low reasoning), minimax m2.5, grok 4.1 fast (no reasoning), and gemini 2.5 flash.

After a little more than 48 hrs of continuous trading, here’s roughly where they stand:

* gpt 5 nano: **+$8700** (down from high of 11.5k yesterday)
* minimax m2.5: +$3100
* gemini 2.5 flash: +$1000
* grok 4.1 fast: -$400

I'm honestly impressed with how gpt5-nano has performed so far, considering it's a relatively cheap model. When I started this I definitely wasn't even expecting it to be in the positives by now. It might just be really good at processing raw financial numbers(idk)? I’m keeping these agents running so we’ll see if these gains stay consistent. Eventually I also want to throw in more expensive models (gpt 5.2, sonnet 4.6) and see how they compete too. 

Also, this project is fully open source so you can check it out and try for yourself! [https://github.com/ryan-yuuu/crypto-daytrading-arena](https://github.com/ryan-yuuu/crypto-daytrading-arena)

**tldr:** gpt-5-nano, good with money??