Correction, October 13, 2026: The AI SDK uses providerOptions.openai.serviceTier; the API request field is service_tier.

Correction, October 11, 2026: Standard-tier fallback applies to requests pinned to unsupported processing regions, such as the EU, rather than all requests from EU visitors.

Vercel AI Gateway now supports OpenAI’s Ultrafast service tier. Vercel says it produces faster output for interactive apps and coding. Requests served at Ultrafast cost six times the standard rate per token, the units of text the model reads and writes.

Requesting Ultrafast and paying for fallback

AI Gateway lets your app request responses from different model providers through one service. Ultrafast supports openai/gpt-6-astra and openai/gpt-6.1-sol. In Vercel’s AI SDK, a library for adding AI features to an app, set providerOptions.openai.serviceTier to "ultrafast". With the Chat Completions or Responses API, the interfaces your app uses to send requests, set service_tier to "ultrafast". If you set nothing, you stay on standard processing.

Two details affect your bill:

  • Region limits. Ultrafast supports US and global processing. A request pinned to an unsupported region, such as the EU, runs at the standard tier.
  • Fallback billing. A request that falls back to another tier is billed at the rate for the tier actually served, not the 6x rate.

For workflows with frequent tool calls, Vercel relays OpenAI’s recommendation to use the Responses API over a persistent WebSocket connection. That keeps the connection open between requests, reducing setup time between turns.

GPT-6.1 Sol launched at GPT-6 Sol’s standard API rates. Ultrafast adds its premium to that model’s standard usage price.

Faster replies need to justify the premium

The premium is easiest to justify when faster output reduces a user’s wait. That includes streamed chat responses and coding steps you are watching. Background jobs need a deadline or another measurable benefit to justify the extra cost.

Compare representative requests on both tiers, recording response time, total cost, and how often requests fall back. Selecting Ultrafast does not give EU-pinned requests a speedup; they receive standard processing and pricing.

Also set a budget on your API key before testing. Vercel says AI Gateway supports budgets for API keys, and a setting that routes all traffic to a 6x tier by mistake would multiply your spend.

End of article