March 2025. Working with Claude. Middle of a task. "You've reached your limit".

Wait 3 hours? No, I switch.

Scheme

Claude ran out - go to GPT. GPT ran out - Gemini. Everything ran out - Openrouter with different models.

Context is lost. Explain from scratch each time. Keep a file with current state - share it in new chat.

Openrouter

Paid, but flexible. Many models, pay per token. When main limits are exhausted - last resort.

Models are weaker, but enough for simple tasks.

How to reduce the problem

• Smaller tasks. Not "write a module", but "write a function".
• Context in a file. Not in your head, not in chat.
• Intermediate commits. Did a part - committed.

How many limits per day

Claude Pro - enough for 4-5 hours of active work. GPT Plus - about the same. Together - almost a day.

If task is big - plan with limits in mind. Complex stuff in the morning on Claude, routine in the evening on whatever's left.