GPT-5.4 mini (medium) is a specialized configuration of OpenAI’s GPT-5.4 mini model, released on March 17, 2026. Designed for high-throughput applications, the "medium" designation refers to a performance tier that balances inference speed with the model's adaptive reasoning capabilities. It belongs to the broader GPT-5.4 family and is optimized for tasks where latency is a critical factor, such as real-time coding assistance, automated data extraction, and agentic subtasks.
The model features a 400,000-token context window, allowing it to process extensive codebases and long-form documents within a single prompt. It is fully multimodal, supporting both text and image inputs natively. A key capability is its advanced support for computer use, which enables the model to interpret screenshots of complex software interfaces and execute multi-step tasks across various digital environments.
In terms of performance, GPT-5.4 mini (medium) provides a significant upgrade over the previous GPT-5 mini, running approximately twice as fast while maintaining higher accuracy. On technical benchmarks, it approaches the performance of the flagship GPT-5.4 model, achieving competitive scores on SWE-Bench Pro and OSWorld-Verified. It is particularly effective for developer workflows, including codebase navigation, debugging loops, and frontend generation.
The model is integrated into the OpenAI API and the Codex platform, where it can act as a subagent to handle less reasoning-intensive work in parallel with larger models. It is also available to ChatGPT users via the "Thinking" feature, which adaptively determines the level of reasoning effort required for a given query. For production workloads, the model offers a cost-effective alternative to frontier-class models while maintaining reliable tool-calling and function-calling accuracy.