So you had generated 2000 lines in 30 minutes and ran out of tokens? What was your prompt?
I’d use a fast model to create a minimal scaffold like gemini fast.
I’d create strict specs using a separate codex or claude subscription to have a generous remaining coding window and would start implementation + some high level tests feature by feature. Running out in 60 minutes is harder if you validate work. Running out in two hours for me is also hard as I keep breaks. With two subs you should be fine for a solid workday of well designed and reviewed system. If you use coderabbit or a separate review tool and feed back the reviews it is again something which doesn’t burn tokens so fast unless fully autonomous.