Issue: OpenAI Flex pricing only partly applied

[davidgilbertson]

Issue you'd like to raise.

In the pricing shown in LangSmith, it appears that for an invocation using the flex tier:

• input flex tokens are treated correctly
• output visible (not reasoning) tokens are treated correctly
• output reasoning tokens are calculated using the full rate, not the flex rate

For the one request where I did the sums it results in LangSmit over-reporting the cost by ~20% (so easy to miss because it's not exactly half or double!)

Suggestion:

I'm not sure if this is an error in logic (looking up the wrong price) or an error in the pricing table. The docs have a link to to a model pricing table - but to view this is says "Upgrade required".

[jacoblee93]

Hey David Gilbertson (@davidgilbertson), are you using our wrapOpenAI integration?

The accounting is a bit tricky but you need to break things down by flex_reasoning and flex in input token details. The built-in wrappers should do this automatically.