This is a partial quote from a blog post.
https://openai.com/ko-KR/index/introducing-gpt-5-4-mini-and-nano/
Today, GPT‑5.4 mini and GPT‑5.4 nano models are being released for ChatGPT. Both models offer the highest performance among the small models introduced so far, and are designed to process large tasks faster and more efficiently while maintaining many of the strengths of GPT‑5.4.
GPT‑5.4 mini provides significantly improved performance at over twice the speed of GPT‑5 mini in areas such as coding, reasoning, multimodal understanding, and tool use. It also achieved performance close to the top-performing GPT‑5.4 model on several benchmark evaluations, including SWE-Bench Pro and OSWorld-Verified.
GPT‑5.4 mini is optimized for tasks where latency directly impacts product experience. This includes coding assistants where fast response is crucial, subagents that quickly process auxiliary tasks, computer usage systems that capture and interpret screenshots, and multimodal applications that infer images in real time. In these environments, the largest model doesn't always mean the best. A model that responds quickly, reliably utilizes tools, and maintains high performance in complex specialized tasks is often more suitable.
GPT‑5.4 nano is the smallest and fastest version of the GPT‑5.4 model, designed for tasks where speed and cost are paramount. It offers significantly improved performance compared to GPT‑5 nano and is well-suited for classification, data extraction, ranking tasks, and subagents handling relatively simple auxiliary tasks. [Omitted]
For ChatGPT, Free and Go plan users can select the 'Think Carefully' option in the + menu to use GPT‑5.4 mini. For all other users, GPT‑5.4 mini will be provided as an alternative model when the usage limit for GPT‑5.4 Thinking is reached. [Omitted]
https://x.com/OpenAI/status/2029620619743219811
