[tosh]

There has been incremental progress for about 1 year from open weight models worse than GPT-3.5 to models in the area of GPT-4.

Same for inference speed/cost: many many incremental improvements within 1 year add up.