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

OpenAI has officially released two new compact AI models, the GPT-5.4 mini and the GPT-5.4 nano, designed to deliver improved speed and efficiency. These models are tailored to manage a high volume of requests at a reduced cost, making them an attractive option for various applications. The GPT-5.4 mini boasts significant enhancements in performance across coding, reasoning, and multimodal tasks, outpacing its predecessor, the GPT-5 mini. According to benchmarks, the mini variant operates at twice the speed and closely matches the capabilities of the larger GPT-5.4 model. On the SWE-Bench Pro, it achieved a score of 54.4%, while the full GPT-5.4 model scored 57.7%. The mini model also excelled on OSWorld-Verified, earning 72.1%, just shy of the GPT-5.4's score of 75%. In contrast, the GPT-5.4 nano is the most compact and cost-effective option in the GPT-5.4 lineup. It is recommended for simpler tasks such as classification, data extraction, ranking, and coding subagents. While it may not match the speed and intelligence of the mini version, it surpasses the GPT-5 mini in coding and tool utilization, with scores of 39.01% on OSWorld-Verified and 52.4% on SWE-Bench Pro. The nano model is not designed for internet browsing or complex reasoning tasks; instead, it is optimized for executing simple functions repeatedly within a larger AI framework. Both the GPT-5.4 mini and nano models excel in coding tasks, including editing, codebase navigation, front-end development, and debugging. The mini variant also features capabilities for reviewing large documents, processing files, and searching through codebases. As for availability, the GPT-5.4 mini can be accessed via the API, Codex, and ChatGPT starting today. In the API, it offers a context window of 400k for a range of tasks, including text and image processing, tool integration, web searches, and file management. The models can also be integrated into applications, command-line interfaces, integrated development environments, and web applications through Codex. Free and Go users of ChatGPT can find the mini model within the “Thinking” option. Meanwhile, the GPT-5.4 nano is exclusively available through the API, designed for simple, high-frequency tasks at minimal costs.