The company stated that GPT-5.4 mini significantly outperforms GPT-5 mini in coding, logical reasoning, multimodal analysis, and tool usage. At the same time, processing speed has more than doubled.

We’re introducing GPT-5.4 mini and nano, our most capable small models yet.
We’re introducing GPT-5.4 mini and nano, our most capable small models yet. X

In benchmarks such as SWE-Bench Pro and OSWorld-Verified, the model approaches the performance level of GPT-5.4.

Source: OpenAI.
Source: OpenAI. 

“GPT-5.4 mini is designed for use cases where response speed directly defines product quality,” OpenAI noted in its announcement.

The company highlighted key scenarios where the new models deliver the most value:

  • AI assistants with near-instant responses;

  • sub-agents executing auxiliary tasks in real time;

  • interface control systems that interpret screenshots;

  • multimodal applications for real-time image processing.

“In these scenarios, the best model is not necessarily the most powerful one. It is the one that delivers instant responses, reliably works with tools, and still handles complex professional tasks,” the company added.

GPT-5.4 nano is the most compact and fastest version in the GPT-5.4 lineup. OpenAI recommends it for classification, data extraction, ranking tasks, and lightweight sub-agents performing simple operations.

Source: OpenAI.
Source: OpenAI. 

Programming

OpenAI emphasized that GPT-5.4 mini is particularly effective in development workflows where rapid iteration is critical. The low-latency model handles targeted code edits, codebase navigation, frontend generation, and debugging cycles efficiently.

Source: OpenAI.
Source: OpenAI. 

“This makes it an excellent choice for tasks requiring both speed and cost efficiency,” the company stated.

In benchmarks, GPT-5.4 mini outperforms GPT-5 mini at comparable response times while delivering significantly faster execution overall.

Computer Use

GPT-5.4 mini demonstrates strong performance in multimodal tasks related to interface control. The model can instantly interpret complex dashboard screenshots and execute actions on a computer with minimal delay.

Access and Pricing

GPT-5.4 mini is available via API, Codex, and ChatGPT. Pricing is set at $0.75 per million input tokens and $4.5 per million output tokens.

GPT-5.4 nano is available exclusively via API. Its pricing is lower, at $0.2 per million input tokens and $1.25 per million output tokens.