OpenAI has introduced GPT-5.4 mini and GPT-5.4 nano, its latest smaller AI models which are said to have been designed to deliver faster responses and improved efficiency for high-volume workloads. The company said that the new models bring several capabilities of the larger GPT-5.4 model to more lightweight systems, with a focus on speed, cost and responsiveness.
The launch follows the recent rollout of GPT-5.4 (referred to as GPT-5.4 Thinking in ChatGPT), which the company positions as its most capable model for professional use.
Multimodal understanding and everyday use
According to OpenAI, GPT-5.4 mini offers improvements over GPT-5 mini across areas such as coding, reasoning, multimodal understanding and tool usage, while also running more than twice as fast. The company said the model approaches the performance of the larger GPT-5.4 model in several evaluations, particularly in coding-related tasks.
GPT-5.4 nano, on the other hand, is positioned as the smallest and most cost-efficient option in the lineup. As per OpenAI, it is intended for simpler and repetitive tasks such as classification, data extraction, ranking and supporting coding workflows.
Both models have been claimed to be designed for scenarios where response time plays a critical role, including coding assistants, systems that interpret screenshots and applications that work with both text and images in real time. OpenAI said in such use cases, faster and more efficient models can often be more practical than larger ones.
Focus on coding, subagents and computer-based tasks
OpenAI highlighted that GPT-5.4 mini performs strongly in coding workflows that require quick iteration, such as editing code, debugging and navigating large codebases. The model is also said to be suited for systems that combine multiple AI models, where larger models handle planning while smaller models like GPT-5.4 mini take on narrower tasks in parallel.
This approach allows developers to build systems where different models work together, improving both speed and efficiency.
The company said that the model is also designed to handle computer-based tasks, including interpreting screenshots of user interfaces and interacting with digital environments. It added that GPT-5.4 mini shows improved performance in such multimodal use cases compared to GPT-5 mini.
Availability
OpenAI said that GPT-5.4 mini is now available across the API, Codex and ChatGPT. In ChatGPT, it can be accessed by Free and Go users through the “Thinking” feature in the “+” menu, while for other users it functions as a rate-limit fallback for GPT-5.4 Thinking.
Within Codex, the model is available across the app, CLI, IDE extension and web, and is designed to handle simpler coding tasks at a lower cost. It can also be used as part of systems where more complex tasks are handled by larger models. GPT-5.4 nano, meanwhile, is available through the API.
