OpenAI unveiled GPT-5.4 mini and nano, compact models with near-flagship performance.

OpenAI is launching compact language models, GPT-5.4 mini and nano, one of which is already available to users of the Free and Go plans. GPT-5.4 mini significantly outperforms GPT-5 mini in areas such as programming, logical reasoning, and multimodal understanding (i.e., processing different types of data (text, images, and voice)). Furthermore, the new AI model is more than twice as fast, and in some tests even approaches the performance of GPT-5.4. OpenAI states that these language models are designed to quickly and efficiently process large volumes of data and are intended for workloads where minimal latency is essential.

Free and Go users can access GPT-5.4 mini via the "Thinking" feature in the ChatGPT menu. For other users, this model serves as a fallback option in case GPT-5.4's query limit is reached. The mini version is better at processing non-text data, such as images and audio, and has a more sophisticated understanding of web search.

Regarding GPT-5.4 nano, the company claims it is the smallest and fastest model, ideal for tasks like classification and data mining, where speed and cost-effectiveness are paramount. It is not available in the ChatGPT service; access to it is provided only through the OpenAI API. The company anticipates that developers will use more advanced models to delegate tasks to AI agents running on GPT-5.4 nano, and this is reflected in the pricing of the new model. The nano version costs $0.20 per million input tokens and $1.25 per million output tokens. The mini version costs $0.75 and $4.50, respectively.

Source