GPT-5.6 Sol Gets Ultrafast: 14x Faster and a New ChatGPT Memory

OpenAI introduced a new Ultrafast tier for GPT-5.6 Sol that allows the model to generate up to 750 output tokens per second, making it up to 14 times faster than standard processing. The speed is achieved in partnership with Cerebras, a company specializing in AI computing infrastructure and wafer-scale architecture. Ultrafast is aimed at real-time applications such as voice, customer support, e-commerce, security, incident response, financial analysis, and research. OpenAI also unveiled Computer History, an opt-in feature for the ChatGPT and Codex desktop app on Mac that uses information from recent user activities to help continue tasks and recognize recurring workflows. Users can control which apps and websites are included, and the feature does not record screenshots, audio, or private browsing sessions. Interaction events are temporarily stored on the Mac and processed in a short-lived Codex session on OpenAI's servers, with event files not stored or used for training and deleted after 48 hours. Computer History will roll out initially for Pro, Business, and Enterprise users, but not in Germany or other parts of the European Economic Area, UK, and Switzerland, with availability promised in the coming weeks.