OpenAI launches its most capable small models yet
What's the story
OpenAI has announced the launch of its new models, GPT-5.4 mini and nano. The former is now available to free ChatGPT users. The company recently unveiled the more powerful GPT-5.4 model for higher-tier users. Now, these new mini and nano variants are being made available, though nano will only be available via the OpenAI API.
Enhanced performance
GPT-5.4 mini approaches the performance of full-sized model
The new GPT-5.4 mini model is said to be over two times faster than its predecessor, the GPT-5 mini. It also approaches the performance of the larger GPT-5.4 model on several evaluations, including SWE-Bench Pro and OSWorld-Verified. The new model brings significant improvements across various tasks such as coding, reasoning, multimodal understanding, and tool/computer use.
Targeted design
Designed for specific workloads where latency is critical
OpenAI has designed these models for specific workloads where latency directly impacts the product experience. This includes coding assistants that need to be responsive, subagents that quickly complete supporting tasks, computer-using systems that capture and interpret screenshots, and multimodal applications capable of real-time reasoning over images. In such scenarios, speed and reliability are more important than sheer size or complexity of the model.
Cost-effective solutions
New models available at lower costs than GPT-5.4
The new models, GPT-5.4 mini and nano, can handle coding workflows like targeted edits, codebase navigation, front-end generation, and debugging loops with low latency. They are available at much lower costs than the more powerful GPT-5.4 model in OpenAI's API, and mini is also available in Codex. The company has also made the new mini model available via its API and Codex platforms for wider accessibility among developers.