Comment by jessenaser

Why does the production model merge vision with the base turbo model such that the output tokens remains 2048 instead of 4096?

Because if we are using just text, why is the extra output size reduced still?