When an assistant is assigned to the "Best ..." language model, it's not properly using the best

[krschacht]

I just tried assigning a newly created assistant to the "Best OpenAI Model" and it got an API error saying "openai-best" is not a recognized model name.

But our code used to be written such that openai-best is a placeholder model name and it gets substituted out for whichever model for this service has best = true. At some point we've broken this. I think this is a nice feature to support because people who are inclined to create many assistants are most likely to want to point them to the "best" one and then when you upgrade your language model to something newly released in the APIs you only have to update a single place.

Anyone want to take this one?
@mattlindsey @duckworth @papayalabs @drnic @voodoo @sergiobayona

[krschacht]

Also, maybe @draknor, @ceicke, @stephan-buckmaster, @jasonpaulso or @htruong if you might want to help with this one?

[stephan-buckmaster]

I can look into this

[stephan-buckmaster]

Looking at the code, when a language model is called "best", should its flags like supports-tools be taken from the "best" (according to the best flag) model?

[krschacht]

Oh interesting: I think if language model named “best” is redirecting to a different model, then we should use the flags from the target model.

Maybe we should disable flags completely for these special models named “best” to avoid that confusion.

[stephan-buckmaster]

Pull-request is #604