50% reduction in cost charged to customers, inference cost may as well be the same, we don't know.
replyRecent reports that OpenAI found optimisations would explain that these are legit
reply"while being more performant"
reply..on some specific set of benchmarks ;)