Research firm OpenAI has released new versions of its GPT-5.4 mini and nano, which are faster and cheaper. These models, capable of handling complex coding tasks and analyzing large datasets, can memorize up to 400,000 tokens of information at a time. They prioritize speed and low operating costs over larger models.

Competition between large companies in this field is intensifying. GPT-5.4 mini mainly competes with Google's Gemini 3.1 Flash-Lite, Anthropic's Claude Haiku 4.5, and Mistral Small 4. Despite its superior performance, Mistral Small 4 is currently the cheapest on the market. Those who want data security prefer Mistral's open source approach.

These new models indicate a shift in the enterprise sector away from simple chatbots and towards AI agents that can perform tasks on their own. It is expected that such small models, which can recognize images and accurately follow complex instructions, will be widely used for business purposes.