Deferred Processing

50% off eligible durable text jobs; pins opt in.

When work can wait, Hydracept uses deferred processing: the cheaper latency-tolerant provider tier, priced at half of standard token rates. Interactive invoke, stream, and ordinary pins stay on standard processing.

Automatic on durable jobs

POST /v1/capabilities/{key}/jobs for text work. No extra execution preference is required.

Pinned Responses, opt in

POST /v1/inference/pinned stays on standard processing. Set processing: deferred on eligible OpenAI Responses pins for service_tier=flex. One logical model execution, no truncation rewrite, no Flex→Standard fallback. Capacity 429s retry inside the admission deadline.

Discover before you submit

GET /v1/capabilities/{key} exposes features.deferredProcessing with discountBps, appliesTo, defaultFor: job_async, and optInFor: pinned.

Honest receipts

The discounted upstream basis is what you are charged. Durable jobs may fall back to standard rates if Flex is rejected. Pins never change the pin: pre-inference 429s retry, Flex→Standard does not.