OpenAI launches GPT-5.4, the most factual and efficient model According to a report by Jinshi, OpenAI has officially launched GPT-5.4, stating that it is the most factual and efficient model - with fewer tokens and faster speed. In ChatGPT, the thinking of GPT-5.4 has improved in deep web research, with better context preservation during long-term thinking - now users can interrupt it, add instructions, or adjust direction during the model response process. The guidance function has been launched on Android and web platforms this week, and the iOS version is about to be released. OpenAI developers claim that GPT-5.4 has native computer usage capabilities. Codex and API support up to 1 million contextual tags, enabling optimal proxy encoding for complex tasks, providing scalable tool search in larger ecosystems, and enabling more efficient inference for long, tool intensive workflows.