GPT-5.5 Instant was announced for the API as chat-latest. Now that the alias resolves to GPT-6 Luna, will OpenAI provide a dedicated, version-pinned GPT-5.5 Instant API model before October 14? Developers need a stable identifier, not a moving alias.
You can use a stable identifier gpt-5.5 and not a moving alias, and with a “none” reasoning effort, you can expect the same non-reasoning:
{
"model": "gpt-5.5-2026-04-23",
"input": [
{
"type": "message",
"role": "user",
"content": [
{
"type": "input_text",
"text": "Hi!"
}
]
}
],
"max_output_tokens": 3000,
"store": false,
"reasoning": {
"effort": "none"
},
"include": [
"reasoning.encrypted_content"
]
}
I would expect gpt-6-terra to have a longer lifetime, if you wish to evaluate that model instead. It is a reduction in price.
Thank you for the suggestion. However, I think this addresses a different problem from the one I’m trying to solve.
Setting reasoning.effort to none on a version-pinned GPT-5.5 model may disable explicit reasoning, but it does not establish that the resulting model is identical to GPT-5.5 Instant. A reasoning configuration is not necessarily a substitute for a distinct model variant.
My request is specifically about preservation, not performance or benchmark results. GPT-5.5 Instant has its own conversational characteristics and behavior that I would like to retain access to, rather than approximate using another model.
Since chat-latest previously provided access to Instant but now routes to a different model, could OpenAI consider publishing a dedicated, version-pinned GPT-5.5 Instant API identifier, even if it is offered as a legacy model with no further updates?
I understand that maintaining older models has infrastructure costs. Even so, a stable identifier and a defined deprecation policy would give developers an opportunity to preserve access to models they depend on.
With the October 14 sunset approaching, clarification on whether this is possible would be greatly appreciated.
Hi @magic_cue_ball,
For clarity, the October 14 retirement notice explicitly excluded API access. Also, The Chat Latest documentation confirms that its underlying model changes regularly and currently lists no dated Instant snapshot.
I’ll surface this to the engineering team so they can review the request and consider whether it could be implemented. I don’t have a release commitment to share.
Sanket