Unintentional Battle: Gemini 2.5 Pro, DeepSeek and ChatGPT 5.1

I happened to have the chance to compare three pretty good neurons in the field.

My wife is engaged in creating schemes for embroidery and threw up a great task. They were given a ready-made macro to calculate the length of threads. But it was impractical to use it: constant «copying», saving, unnecessary clicks.

We talked it over, I got the sources and got down to business. I said the usual:

«What's there to do? 5-10 minutes and it will be ready...»

Yeah, sure. Took me an hour and a half.

The essence of the problem

I wrote one prompt and threw it into all three neurons at once. All three of them happily produced ready code. But here's the problem. none of the options worked..

The main problem was the format on entry - it was a .RTF. No one could handle him properly.

Gemini 2.5 Pro. Constantly feeding me excuses:
- «Ah, I see. Thanks for the screenshot and file, that helped a lot!»
- «I get it. It's some kind of mysticism. We've tried everything...»

A few messages later, he said that it was impossible to do it in the browser. We started «pythonizing», but nothing came out.

DeepSeek. It immediately started generating code, but, unfortunately, it was not working. Neuronka could not read the file correctly.

ChatGPT 5.1. In the end, he was the one who saved me. I read the code and generated the page. All that was left was to fine-tune the little things, which I had already done through Gemini.

It was ChatGPT 5.1 that was able to solve the turnkey problem.

Conclusion

What's my point? I am increasingly convinced that it is advisable to throw in the task to multiple neurons at once. And then either take the winner's finished result, or combine their reasoning and remake the prompt.