OpenAI introduces GPT‑5.4 mini and nano, faster and smarter small AI models

OpenAI has launched two new smaller AI models, the GPT-5.4 mini and GPT-5.4 nano. These are lighter and faster models, which are designed for handling large numbers of requests at lower costs. Here's everything you need to know about the enw GPT 5.4 series models.

GPT‑5.4 mini and nano: Features and benchmarks

The GPT‑5.4 mini comes with improved performance across coding, reasoning, multimodal tasks, and tool use, compared to the GPT-5 mini model. The company also stated that it runs 2x faster and performs close to the larger GPT-5.4 model.

In terms of benchmark scores, the GPT-5.4 mini scored 54.4%, whereas the GPT-5.4 scored 57.7% on SWE-Bench Pro. On OSWorld-Verified, it reached 72.1%, closely trailing the GPT 5.4’s 75% score.

On the other hand, the GPT‑5.4 Nano is the smallest and cheapest version of GPT-5.4. The company recommends the model for simpler tasks such as classification, data extraction, ranking, and coding subagents.

While it does not perform as fast and smartly as the GPT 5.4 mini, it beats GPT-5 mini in coding and tool use. The model scored 39.01% on OSWorld-Verified and 52.4% on the SWE-Bench Pro. Hence, it is not built to browse the internet or to manage complex multi-step reasoning tasks. It is built for a kind of task that can run thousands of times a day in the background of a larger AI system.

Overall, the GPT‑5.4 mini and GPT‑5.4 nano can be used for Coding-related tasks such as edits, codebase navigation, front-end generation, and debugging, as both surpases the GPT 5 mini models.

GPT‑5.4 mini also comes with ability to reviewing large files, processing docs, and even search codebases. Lastly, the models can also be used for computer and multimodal tasks.

GPT‑5.4 mini and nano availability

The GPT‑5.4 mini is available in the API, Codex, and ChatGPT, starting today. 

In API, the GPT-5.4 mini offers a 400k context window for tasks like text, images, tools, web search, files, and even computer actions.

In Codex, the models can be integrated across apps, CLI, IDEs, and web, and can manage simpler coding tasks.

In ChatGPT, it's available for Free and Go users within the “Thinking” option.

The GPT-5.4 nano is only available via API, for simple, high-volume tasks at very low cost.