OpenAI has reinforced its catalog of artificial intelligence models with the launch of GPT-5.4 Mini y GPT-5.4 Nanotwo new options designed for high-volume tasks that seek to compete with solutions like Google’s Gemini 3 Flash.
Since the launch of GPT-5 last summer, the company had focused its advances on more powerful versions such as GPT-5.4, leaving the evolution of its compact models in the background. This allowed other competitors to gain ground in the segment of fast and efficient solutions. Now, OpenAI rebalances its offering with two models that prioritize speed and cost.
On the one hand, GPT-5.4 Mini It is positioned as the company’s most advanced small model to date. It significantly improves GPT-5 Mini in tasks such as programming, reasoning, multimodal understanding and tool use, in addition to offering more than two times the speed of performance. Even in different benchmark tests, this model comes close to the performance of the GPT-5.4 itself.
For its part, GPT-5.4 Nano It is presented as a cheaper alternative, aimed at tasks such as classification, data extraction or code agents, clearly surpassing its predecessor in performance.
In terms of availability, GPT-5.4 Mini can now be used in ChatGPT, Codex and through the developer API. It offers support for text and images, as well as advanced features such as using tools, web searching, file management, and running in computing environments. Besides, It has a context window of 400,000 tokens. Its price stands at $0.75 per million input tokens and $4.50 per million output tokens.
For developers, this model is especially attractive as it only consumes 30% of the GPT-5.4 quota in tools like Codex, making it suitable for less demanding tasks. On ChatGPT, it is available for free and Go users via the “Think” feature, while on premium plans it acts as an alternative when the main model is not available.
Instead, GPT-5.4 Nano is reserved exclusively for the APIwith a lower cost of $0.20 per million tokens for input and $1.25 per million for output.