I'm still evaluating at the moment, I'm using a targeted model approach - one for the conversation layer, another for coding etc.

Currently working on a memory system to help the model changes, so it's all a bit up in the air for now haha.