Get Request
One request by id, owner-only. Same item shape (and the same billingStatus failure semantics) as the list.
Authorizations
Use the format: Key modelrunner_key
Path Parameters
Response
The request.
One inference request as returned by the list/detail/stream surfaces. IMPORTANT: status alone does not signal success — a failed generation is normalized to status: COMPLETED with billingStatus: 'failed'. Always read billingStatus (and error) on these surfaces.
The owner/alias endpoint that served the request (after any provider fallback).
What the request ran against. Absent on older cached payloads; treat a missing value as model.
model, wrapper, deployment The Serverless GPU deployment this job ran on. Null unless targetKind is deployment.
The endpoint originally requested, when a fallback served it. Null/absent otherwise.
IN_QUEUE, IN_PROGRESS, COMPLETED, FAILED, CANCELLED The success discriminator: 'failed' marks a failed generation even when status is COMPLETED.
pending, partial, charged, failed Failure detail, when the generation failed.
The stored client input. {} after payload retention removed it — see payloadsPurgedAt.
The model output. {} after payload retention removed it — see payloadsPurgedAt.
Set when input/output are empty because retention removed them, not because the model returned nothing.
What this request actually charged, as a decimal string.
Owner-controlled publication state; requests default to private.
The caller-supplied metadata tags from submit time.

