OpenAI is quietly building toward a broader release of “Ultrafast,” a faster paid tier for its developer API, according to TestingCatalog, which found a hidden speed selector inside the Responses API Playground offering three choices: Standard, Fast, and Ultrafast.
OpenAI first previewed Ultrafast alongside GPT-5.6 Sol. The company says the tier can push throughput to roughly 750 tokens each second, near 14 times the pace of Standard, running on chips from Cerebras. Access today is limited to select customers.
The timing lines up with OpenAI’s DevDay on September 29. TestingCatalog also spotted new account tiers, including one called Accelerate aimed at production workloads, suggesting the company plans to sell speed itself as a product feature rather than bundle it into a single API price.
It is not yet confirmed whether Ultrafast will extend to the newer GPT-6 Sol and GPT-6 Astra models. Developers building latency-sensitive products, real-time agents or live customer support tools, should watch DevDay for pricing details before committing infrastructure budget to a slower tier.
Reported by TestingCatalog (Alexey Shabanov) on 26 September 2026.
