OpenAI has introduced a limited preview of Ultrafast mode for GPT-5.6 Sol, saying the new tier can deliver up to 14 times faster output than the model’s standard version and reach 750 output tokens per second versus a roughly 53 tokens per second baseline. Announced on August 13, the offering is initially available to a select group of customers through the OpenAI API, with broader access planned as computing capacity expands. The service builds on OpenAI’s existing Fast mode, which speeds GPT-5.6 Sol by up to 2.5 times, but Ultrafast is positioned as a separate tier enabled by a partnership with Cerebras and its wafer-scale chip architecture. OpenAI and Cerebras said GPT-5.6 Sol in Ultrafast mode completed Humanity’s Last Exam, a 2,500-question benchmark, in 11 hours and 11 minutes, compared with 78 hours and 27 minutes for Anthropic’s Claude Fable 5. OpenAI also said Ultrafast delivered a 5.6x end-to-end speed improvement on GDP-Val without quality degradation. The companies said the model runs about 11 times faster than Claude Fable 5 and roughly 5 times faster than Opus 4.8 in Fast mode. Early users include Jane Street, Basis, Rogo and Podium, with OpenAI highlighting applications in incident response, financial research, security operations, voice AI, e-commerce, and research workflows where lower latency can materially change how the model is used.