Runtime
4 current practices, and 2 retired.
StandardRuntimeRN-001Pin a stable prefix so the cache actually hitsPut system prompt, tools and few-shots ahead of anything that varies per request.
Reviewed Aug 2026RuntimeRN-002Route between models by task complexitySend high-volume easy work to a cheap small model and escalate only what's genuinely hard.
Reviewed Aug 2026StandardRuntimeRN-003Constrain generation to a schema instead of parsing free textUse native structured outputs so the response always parses and conforms.
Reviewed Aug 2026RuntimeRN-004Batch anything that isn't latency-sensitiveUse batch APIs for offline and bulk work.Reviewed Aug 2026
Reviewed Aug 2026RuntimeRN-002Route between models by task complexitySend high-volume easy work to a cheap small model and escalate only what's genuinely hard.Reviewed Aug 2026StandardRuntimeRN-003Constrain generation to a schema instead of parsing free textUse native structured outputs so the response always parses and conforms.Reviewed Aug 2026RuntimeRN-004Batch anything that isn't latency-sensitiveUse batch APIs for offline and bulk work.Reviewed Aug 2026