If you use Claude, GPT, and Gemini, how do you split your day-to-day tasks between them?

I’m planning my API budget and would like to hear how people actually use multiple models.

Do you mostly stick with one provider, or pick different models for different tasks? For coding, long documents, and information extraction, what matters most to you: output quality, response time, or cost?

I’d especially like to hear from people who’ve used them regularly. Have you switched models for a particular task? What made you switch?

Concrete examples would be helpful as I work out how to allocate my API budget.