Because:

1. GPT4 is on a different scale to CodeLlama

and

2. If open models do catch up, they will have moved on to GPT5 or GPT6 and the gap will have widened again