Hi Team ,
what is finite limits for GPT-5.6 family for low and high.
Currently, “high” is still 2500 tokens on gpt-5.6.
“low” has been continuously non-functional on gpt-5 models despite the documentation.
It is just the language of documentation that is ambiguous.
It may be left ambiguous because now with no dated snapshot, OpenAI can change the model’s “efficiency”, performance, features, or quality on API developers with no disclosure or notice (like was already done against “snapshots” before, but not to a great degree) - or pricing.
I’ve done extremely rigorous checks against different resolutions, even where OpenAI documentation or algorithmic method is ambiguous, wrong, or non-existent, to deliver an online cost and “patches” calculator: