Quality based routing

[JustinGuese]

How about we would offer a "quality based routing", like instead of setting the model, one would set something like

• good
• medium
• fast

and in the backend it will map to try the models in that order, e.g.

{
    "good": [
        "gpt-4o",
        "gpt-4",
        "gpt-4-turbo",
        "claude-3-sonnet",
        "gemini-pro",
        "claude-3-opus"
    ],
    "medium": [
        "llama3-70b",
        "mixtral-8x22b",
        "mixtral-8x7b",
        "claude-3-haiku",
        "command-r+"
    ],
    "fast": [
        "gpt-3.5",
        "gpt-3.5-turbo",
        "command-r"
     ]
}

Why?
Because I'd love to try gpt-4o first, but it's not always available or I am reaching a rate limit. If it's not available I want to try 4, 4-turbo and so on.

Same for if I just want a quick result, where I'd be willing to have the same done for 3.5-turbo and so on....

Currently one would have to implement the logic in their own app, like try 4o, if it doesn't work, try 4 and so on. Would save me a lot of work ;)

What do you think?

[JustinGuese]

An even simpler implementation would be setting the model to "auto", and then gpt4free automatically tries all from best to worst, according to e.g.
https://artificialanalysis.ai/leaderboards/models