OpenAI has released GPT-5.4 mini and GPT-5.4 nano, two optimized versions of its main model. Designed for speed and reduced cost, they maintain solid performance in code and reasoning. The main novelty is the integration of GPT-5.4 mini into ChatGPT for free and basic users, accessible via the Thinking option, which democratizes access to advanced capabilities.
Technical performance and applications in development 🧪
In benchmarks like SWE-Bench Pro and OSWorld-Verified, GPT-5.4 mini's results approach those of the full model. Its architecture, optimized for efficiency, achieves more than double the speed of its predecessor. For subscribers, it acts as a backup when usage limits are exceeded, ensuring continuity in practical tasks without noticeable interruptions in response quality.
The "economical" model that thinks for you (but fast) âš¡
With the arrival of the mini, we can now have an assistant that reasons at full speed, ideal for when patience is scarce. It's as if the main model had been put on a strict diet: it does almost the same, but without so many resources. That said, don't tell the big GPT-5.4 that its little brother does the work for a fraction of the cost.