GPT-5.4 nano (medium) is a high-efficiency, low-latency language model released by OpenAI on March 17, 2026. Positioned as the most cost-effective variant in the GPT-5.4 family, it is specifically designed for high-volume workloads and real-time applications where speed and operational budget are the primary constraints. The model is part of a broader shift toward "agentic delegation," where it serves as a lightweight sub-agent for flagship models, handling focused tasks like classification, ranking, and data extraction.
Technically, the model supports multimodal inputs, enabling it to process and reason over both text and images. It features a substantial 400,000-token context window, allowing for the processing of extensive codebase segments or long-form documentation. Despite its small footprint, it demonstrates significant proficiency in coding and structured reasoning, scoring 52.4% on the SWE-Bench Pro evaluation and 82.8% on GPQA Diamond.
OpenAI has optimized the model for the "subagent era," recommending its use for codebase navigation, file reviews, and parallelized subtasks. It is engineered to provide a responsive user experience in interactive systems, such as coding assistants and computer-using agents that require frequent screenshot interpretation. At its release, the model was priced at $0.20 per 1 million input tokens and $1.25 per 1 million output tokens, making it OpenAI's most affordable offering to date.