I get GPT 4.1 as default model

[isidornikolic]

Works fine in Stable. I can only repro in Insiders.

1. code-insiders --transient
2. You get the chat welcome view. Send a message "what is 2+2"
3. Login with a paid account (my MSFT joined gh account)
4. You get GPT 4.1 as default model 🐛

I can not 100% repro. So might it be a timing issue?

cc SteVen Batten (@sbatten) Johannes Rieken (@jrieken) Benjamin Pasero (@bpasero)

[bpasero]

Isidor Nikolic (@isidorn) for me it just remains as "GPT 5 mini" 🤔 . Thats the model I had when I signed in with my private GH account.

Can we just cut this short by picking "Auto", I think this is just a mess to deal with otherwise.

[isidorn]

Yeah - I am fine if we just pick "Auto".

Though let's see if Logan has concerns.

[lramos15]

Is the proposal that we get rid of default model experiments and just do auto?

[isidorn]

I might have misunderstood - I thought the proposal is to just pick Auto when we are going out of the welcome view due to weird timings.

We can not yet get rid of the default model experiment - we might be able to do that next year.

[bpasero]

I think every SKU has like a default model right? Then I guess we should automatically pick that default model per SKU if and only if its considered to be more premium.

[lramos15]

I think we should just wait until the default model experiment times out since we will have auto everywhere by end of year which will simplify this.

Isidor Nikolic (@isidorn) Why can't we make this move now? What is needed default wise that we cant use auto for?

[isidorn]

Isidor Nikolic (@isidorn) Why can't we make this move now? What is needed default wise that we cant use auto for?

1. Auto needs to be GA (happening in December)
2. Auto needs to provide more value to the users: dynamic model selection based on capacity
3. Auto needs to be able to give models that are "unconfigured"

You can check the auto-model-team channel as Nhu Do (@nhu-do) shared steps / plans for making Auto as default for all.