OpenAI has just introduced two new models, GPT-5.4 mini and GPT-5.4 nano, designed to optimize processing speed and cost for large-scale artificial intelligence systems, while still inheriting many capabilities from the high-end version GPT-5.4.
This is the next step after the company launched GPT-5.4, a prominent version with flexible feedback direction adjustment, improved in-depth web research and enhanced reasoning with long context.
Among them, the GPT-5.4 mini is considered a significant upgrade compared to the previous GPT-5 mini.
This model significantly improves programming, reasoning, multi-modal understanding and tool use capabilities.
Notably, the processing speed of the GPT-5.4 mini is twice as fast, while operating costs are significantly lower.
According to performance tests, this model achieves a task completion rate close to GPT-5.4, but is more optimized in terms of latency and cost, especially suitable for large-scale programming processes.
In the lower segment, GPT-5.4 nano is designed for priority speed and cost tasks such as data classification, information extraction, content ratings and simple encoding problems.
This is a suitable choice for systems that need to handle large volumes with near-immediate response requirements.
The new models focus on scenarios where latency directly affects the user experience.
For example, programming assistants need quick response, AI assistants handle parallel tasks, or multi-mode applications need real-time image analysis.
A notable point is the ability to deploy in a multi-model system. In the Codex platform, GPT-5.4 can play a coordinating role, dividing small tasks such as source code search, document reading or file analysis for parallel processing mini-models.
Regarding integration capabilities, GPT-5.4 mini is present in API, Codex and ChatGPT. The model supports text and image input, function calling, web search, file processing and has a context window of up to 400,000 tokens.
On ChatGPT, Free and Go users can access through the "Thinking" option, and this is also a backup option when GPT-5.4 reaches a speed limit.
Meanwhile, GPT-5.4 nano is currently only provided via API, targeting businesses that need to deploy AI at the lowest cost.
In terms of price, GPT-5.4 mini costs 0.75 USD per million tokens input and 4.5 USD for output.
With GPT-4 nano, the cost is 0.2 USD and 1.25 USD respectively, showing OpenAI's clear strategy in expanding AI access to more audiences.
