OpenAI has announced Ultrafast, a new technology developed in partnership with AI chip manufacturer Cerebras that dramatically accelerates the GPT-5.6 Sol model. The innovation enables the model to generate up to 750 tokens per second—14 times faster than the traditional version and approximately 11 times faster than Claude Fable 5.
The feature is currently available via API to a limited group of customers, with broader expansion planned as capacity increases. According to OpenAI's official site, this represents a significant leap in inference speed that could reshape how developers and enterprises deploy large language models in production environments where latency matters.