OpenAI Previews 'Ultrafast' Mode for GPT-5.6 Sol — Up to 14x Faster via Cerebras

In one line: OpenAI and Cerebras are previewing an "Ultrafast" API tier that runs GPT-5.6 Sol at up to 750 output tokens per second — 14 times faster than the standard tier.

Key points

• The Ultrafast tier targets 750 output tokens per second, powered by Cerebras silicon, which OpenAI positions as up to 14× its Standard tier.
• It is in limited API preview for a select group of customers; no price, no GA date, and no model ID string have been published yet.
• Initial use cases being tested include coding assistance, finance, customer support, research, and incident response.
• All performance figures are OpenAI-measured; no independent third-party benchmark exists yet.

Why it matters

Latency is one of the hardest constraints for real-time AI agents and interactive services. If the 14× speed claim holds up under independent testing, it opens the door for AI adoption in domains that have been locked out by lag — live customer interactions, code autocomplete, and real-time analytics. The missing price tag, however, means total cost of ownership is still unknown.

Read more

• OpenAI introduces 'Ultrafast' mode for GPT-5.6 Sol — TechCrunch
• Previewing Ultrafast mode (official) — OpenAI