I cancelled my subscription two days ago. I was a customer since October last year. I could get a decent bit of work done on just the 20$ subscription, but since this monday, I can barely get two prompts in before hitting my limit.

Same codebase, same sort of prompt, same scale. I was already on the fence. Models like Qwen, Kimi, or GLM5 already go a very long way while being vastly cheaper, and the new openAI models feels equivalent but with higher limits.

This is getting to the point where the right harness makes a bigger difference than the right model. I've been experimenting with some planner-executor-reviewer setup in opencode, and I'm starting to feel like multiple smaller models working together are netting me better results.