> not for working on real codebases You don't pick just one model to "work on real codebases". You use a very advanced model to plan, and a not-very-advanced,

> not for working on real codebases You don't pick just one model to "work on real codebases". You use a very advanced model to plan, and a not-very-advanced, cheaper, faster model to execute planned tasks. This saves money and speeds up work. This is the guidance from Anthropic & OpenAI.