OpenAI replaces Priority Processing with Fast mode, offering 2.5x speed on GPT-5.6 Sol

OpenAI is retiring Priority Processing in favor of a new API tier called Fast mode, announced in the same July 30 update that cut prices on GPT-5.6 Luna and Terra. For GPT-5.6 Sol, the company's top reasoning model, Fast mode is the direct upgrade path for developers who need lower latency and are willing to pay for it.

What's new

"We're also introducing Fast mode in the API, which replaces our Priority Processing offering," OpenAI said in its API changelog. "For GPT-5.6 Sol, Fast mode now delivers up to 2.5× faster speeds than standard processing at twice the price. This change is backward compatible: requests tagged priority will automatically use Fast mode."

That backward-compatibility detail matters operationally: any application already passing a priority flag to the API keeps working without code changes, and simply starts getting routed to the new, faster tier at the same 2x price point Priority Processing charged. The change applies across both of OpenAI's current API surfaces, v1/responses and v1/chat/completions.

The Fast mode launch arrived bundled with pricing cuts for the rest of the GPT-5.6 family — Luna dropped 80% and Terra dropped 20% — meaning the same changelog entry touched both ends of OpenAI's price-performance range: cheaper low-cost inference on Luna and Terra, and a clearer, faster premium tier for Sol.

Context

Priority Processing was OpenAI's existing mechanism for developers who needed guaranteed low latency for production workloads and were willing to pay a premium over standard throughput-oriented serving. Renaming and re-architecting it as "Fast mode," rather than simply raising Priority Processing's performance, signals OpenAI is standardizing the concept as a first-class serving tier across its model lineup rather than a one-off option bolted onto specific models. The move comes one day after OpenAI shipped GPT Transcribe and GPT Live Transcribe, and in the same week the company published research claiming it used GPT-5.6 to cut its own serving costs 20% and lift token efficiency 15% — a pattern of OpenAI treating inference-serving efficiency itself as a product surface, not just a backend optimization.

Why it matters

For developers building latency-sensitive products — voice agents, coding assistants, anything in a human-in-the-loop interaction loop — a documented, named "Fast mode" with a fixed price-performance tradeoff (2.5x speed at 2x price) is easier to reason about and budget for than an undocumented "priority" flag. The backward-compatible migration also means OpenAI can retire the older mechanism without breaking existing integrations, a low-friction way to consolidate its API surface. Combined with the same-day Luna and Terra price cuts, the update reads as OpenAI tuning the GPT-5.6 lineup's price-performance curve across its full range, rather than treating pricing and latency as separate levers.