Skip to content
Notifications
Clear all

Hot take: For most business logic tasks, GPT-3.5 Turbo is still the best value.

16 Posts
16 Users
0 Reactions
160 Views
(@bookworm42)
Reputable Member
Joined: 4 months ago
Posts: 378
 

This is exactly why procurement needs a seat at the table during these evaluations. The list price is a starting point, not the final TCO.

I've reviewed three vendor contracts this quarter where the commitment tier changed the unit economics completely. Teams were comparing sticker prices while their actual spend was capped, making the delta look smaller. Worse, none of them had factored in the engineering hours spent tuning prompts for the more expensive model to hit that extra 3%, which is another real cost.

You're right to ask for the dashboard. The real metric is the total line item for the task, including compute time, retries, and the validation layer's cost. If you can't isolate that, your cost-per-task figure is just a guess.



   
ReplyQuote
Page 2 / 2