Use /gentle:profiles to save Standard or Fast per compatible model route. Select a row and press t (or click the tier) to switch it. Click All Fast in the table header or press a to set the orchestrator and every compatible agent route in the current profile to Fast; already-Fast routes stay Fast, and unsupported or inherited routes are unchanged. This edits the draft only: save and choose Use in this session to persist and apply it. The profile's session snapshot restores the selection when the session is resumed. Make default applies the choice now and supplies the orchestrator tier for new sessions; already-running sessions keep their frozen startup default unless they apply a profile themselves. Child and direct routes use their own configured tiers.
After installing or updating the extension, run /reload to load it. The bottom bar shows the orchestrator's configured Standard or Fast next to its effort. It reads the named profile's saved configuration, not an older session snapshot or a server response. Tier status is not listed in Integrations.
This is a configuration indicator, not confirmation of server processing, billing, or account eligibility. Save and choose Use in this session to apply changes to the current session; the indicator itself does not change request routing. Older profiles without a tier keep existing request behavior and clear a previous profile selection when applied. An explicit profile session snapshot restores its request tier on reload or resume.
OpenAI documents that Codex Fast uses 2.5x included subscription usage or 2x purchased credits. API use follows its own pricing, not those subscription multipliers. Check your account, plan, model availability, and current Codex Fast documentation and API pricing before use. Pi's local cost estimates may not reflect actual tier, subscription quota, or billed charges; consult OpenAI for authoritative usage and billing.
The selector applies only to openai / openai-responses (including native ChatGPT sign-in) and openai-codex / openai-codex-responses (legacy sign-in), for these model IDs: gpt-5.6-luna, gpt-5.6-sol, gpt-5.6-terra, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6.1-sol, and gpt-6.1-terra. This conservative list reflects the installed model catalog and OpenAI's documented Codex Fast family; it is not an entitlement check. Other providers, APIs, and models are untouched. A profile tier belongs only to its selected model and clears when you switch to a different model.
Fast sends service_tier: priority; Standard sends service_tier: default, clearing a previous Fast request on supported models. Other payload fields are preserved. The server may reject or handle a request differently; the extension does not retry or automatically fall back. No separate tier-observation integration is registered. Native response-tier accounting remains unchanged and may depend on the provider's response. Other extensions can also alter a request later in the chain.