I can do a test a little later today and share the results, but my guess would be that nothing will happen related to the behaviour I mentioned. What’s your hypothesis, Curt?
I don’t think it influences the model, but not sure where this parameter comes into play. So test and make sure.
I know that GPT-4-Turbo is really long winded compared to GPT-4. But haven’t done much with the GPT-3.5 version you are working with.
Worst case use 4-Turbo 
Assuming you want longer responses. In general, I don’t like them.
1 Like
For the projects I’m working on now, waiting times of gpt-4 family are unfortunately restrictively long.
1 Like
From my tests GPT-4-Turbo is 20-40% faster than vanilla GPT-4.
But GPT-4-32k is king and is 100% faster than GPT-4.
1 Like
How did you get access to GPT-4-32k, man?
I could access it through OpenRouter obviously but I’m reluctant to use it in production.
I don’t know exactly. Probably for being a forum Regular.
1 Like
@curt.kennedy I just did this test, same results as expected.
1 Like