OpenAI NewsAug 13, 2026, 18:00 UTC+8

OpenAI previews Ultrafast mode: GPT-5.6 Sol at up to 14X the speed

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed

OpenAI added an Ultrafast inference tier for GPT-5.6 Sol, running on Cerebras chips at up to 750 output tokens per second—14× faster than standard. The preview launches via the API first, targeting latency-sensitive workflows like incident response, financial research, and real-time customer support. OpenAI’s own teams are using it for on-call debugging and to tighten overnight research loops into same-day iterations. The post does not disclose pricing or a general release date; access is by application only.

Why it matters: OpenAI's Ultrafast preview pushes GPT-5.6 Sol to 14X standard speed via Cerebras silicon, with three concrete latency-sensitive use cases. No pricing or GA date disclosed, capping the score at 82 rather than pushing into the must-write-same-day band.

Read the original ↗Export Markdown

Take

The reason to click: this ties speed and intelligence together for the first time at the frontier tier. Until now, real-time latency meant dropping to a smaller model. OpenAI took GPT-5.6 Sol—their strongest model—and ran it on Cerebras wafer-scale chips, hitting 750 tok/s. They're already using it internally for on-call debugging and compressing overnight research loops into same-day sessions.

I'd discount two things. One, no pricing is disclosed—just an application form. Cerebras hardware isn't cheap, so this speed tier probably won't be either. Two, it's API-only for now; no ChatGPT integration, no general-availability date. The Jane Street and Podium quotes are real early feedback, but the sample is small.

If the pricing lands right, this matters for latency-sensitive workflows: trading, live support, incident response. But don't read it as "everything should run Ultrafast"—most async tasks are fine on standard mode and cost less.

Source: OpenAI News · openai.com
Published: Aug 13, 2026, 18:00 UTC+8 · Aug 13, 2026