What happened?

In one of my recent PRs (#4603), while removing deprecated models from the openai-codex provider, i foolishly thought that i would also add gpt fast models by creating new model entries with ids such as gpt-5.5-fast, thinking it would work. Unfortunately that's not the case.

Fast models are just the base model id (eg. gpt-5.5) called with service_tier: "priority".

Steps to reproduce

1. Use GPT 5.5 Fast

Expected behavior

I'm opening this issue because, before sending a PR to fix my own mess, i think this is a broader product choice that should be discussed with the maintainers. In particular, the two possibilities for handling this are:

1. Fast models should be their own models with their own id, but have a mapping that points to the base model and the priority level for when we build a request (this is the option i'm leaning toward).
2. Models that have support for fast mode should have a picker (like for the thinking level) to select if we want 'Fast' or 'Default'.

As i said, i'm open to fix this myself as soon as i have your final decision

Version

0.75.0