Priority Became Fast Mode — What Changed

Fast mode and Priority processing are one service tier under two names. OpenAI renamed it on July 30, 2026, and both service_tier: "fast" and service_tier: "priority" are still accepted.

What changed besides the name

This page originally said the rename changed nothing else. That was wrong: OpenAI's own Fast-mode guide says the same change made Fast up to 2.5x faster than Standard for GPT-5.6 Sol, at a premium over Sol's Standard rate. The same day, GPT-5.6 Luna and Terra were repriced downward. A rename that ships alongside a speed change and a price change is exactly the kind of release where "only the name moved" is the comfortable, wrong reading.

The leftovers of the old name

The response still says priority for GPT-5.6 and earlier models, whichever value you sent — so a script that checks the response's service_tier field against "fast" will report a mismatch on those models. Ramp traffic too quickly in Fast mode and some requests drop to standard speed and standard price, with the response reporting default.

What to use going forward

New code should send fast, and should read the response field rather than assume it. Inside Codex, /fast toggles the tier for the current model.

Guarantees differ by model

The name is shared across models; the terms aren't. Fast for GPT-6 Astra carries no latency SLA, while Fast on GPT-5.6 and earlier models gets the same service-level treatment as Scale Tier. Fast is also billed separately from Scale Tier: its requests don't draw down purchased Scale Tier bundles, and Scale Tier spillover doesn't move into Fast automatically.

Not offered everywhere

The tier isn't published for every model: the Pro models and GPT-5.4 Nano have no Fast row at all. And it is no longer the top speed tier — Ultrafast sits above it, for GPT-6 Astra and GPT-6.1 Sol only. The table above shows each model's Fast input price and its multiple of Standard.