GPT-4 Models have large differences in price

[Timothy Blumberg]

I think that we should allow some further customization of which models that a user actually wants to use when translating, because the regular gpt-4 model is significantly more expensive than the turbo versions of the same models.

But all of these models are grouped together as choices to pull from when translating content. I ran a short test of 75 paragraphs and the price disparity is pretty shocking:

Vanilla gpt4

Turbo gpt4

Proposed solution

I think that it would be a good idea to expose gpt4-turbo as well as all-gpt4 as model options and include some documentation that explains this, because if I wasn't paying attention I could have accidentally significantly overpaid.

Another option would be just to remove vanilla gpt-4 by default since the turbo models have incredibly high limits on even level 1 accounts now:

I'm open to discussing better options here, but I do think that at the very least some documentation updates should be made to make it clear that both turbo and regular models are grouped together like this. Happy to hop in and help if there is some consensus on a good structure for this.

I would just hop in and write the code, but probably worth some discussion since this would involve some changes to the external API.

[yihong]

yes you are right we can fix one specific model, PR welcome for that.

Thank you very much

[Timothy Blumberg]

Perfect. I will take a look at adding this code tonight.

[Timothy Blumberg]

Resolved with that merge