Support gpt-realtime-2

[IngLP]

Feature Type

Would make my life easier

Feature Description

A new GPT Realtime model has been published, gpt-realtime-2.
Any ETA for the support in LiveKit?

⚠️🚨 EDIT: The model is NOT supported, even if it seems to work when specifying as model "gpt-realtime-2". This is because LiveKit does NOT support preambles:
"expected to receive only one message generation from the realtime API"

Thanks!

Workarounds / Alternatives

No response

Additional Context

No response

[schedawg74]

it's already supported, however the addition of reasoning_effort would be nice. or to disable reasoning alltogether.
see https://community.livekit.io/t/gpt-realtime-2-set-reasoning-effort-to-none-or-very-low/1062

[IngLP]

@schedawg74 the model is still not listed in the model literal type. You can force it, yes.
And yes, the biggest need is to be able to pass to the model the reasoning effort.

[IngLP]

⚠️⚠️ The model is NOT fully supported, intermediate generations, gpt-5 style, are NOT supported:
"expected to receive only one message generation from the realtime API"

[SalehHindi]

Yeah honestly this feature was half baked, the preamble support isn't working. The OpenAI Realtime model is generating TWO separate response items for a single user turn, but LiveKit only plays audio for the first one.

[tinalenguyen]

hi everyone, we are working on it in this PR: #5690

please feel free to test it out from that branch, i will update here again once it's merged/released!

[alialfaile10-cloud]

what we need to do now ? can we use it gpt-realtime-2 ?

[jmslau]

It seems like back-channeling and interruption handling isn't great with the current gpt-realtime-2 support. Is that a known issue?