OpenAI Previews Ultrafast Mode: GPT-5.6 Sol at 14x Speed

OpenAI has unveiled a new operating mode for its GPT-5.6 Sol model: “Ultrafast” delivers responses at up to 14 times the standard speed – up to 750 output tokens per second. The mode is in preview and initially available only to a small group of customers.

The speed jump is powered by a partnership with Cerebras. The chipmaker, known for its wafer-scale processors, confirmed the collaboration on August 13 in its own announcement. Cerebras systems specialize in extremely fast inference, competing directly with Nvidia-based setups.

OpenAI frames Ultrafast as an answer to a familiar dilemma: until now, getting real-time speed usually meant settling for a smaller or specialized model. “Ultrafast points to progress in a new direction: more useful work per second,” the company says – full model quality at real-time pace.

Target applications include incident response in IT operations, customer service, financial market analysis and e-commerce – scenarios where latency directly determines usefulness. For agentic workflows that chain many steps, every speed gain compounds.

Open questions remain: OpenAI has not announced pricing, and expansion beyond the small customer group will come only “as capacity grows.” How much Cerebras hardware actually sits behind the offering is not public.