Outlook Business Desk
OpenAI unveiled GPT-5.4 mini and GPT-5.4 nano on March 17, adding to its compact AI model range. These models prioritise faster performance, reduced costs, and efficient handling of large-scale tasks across coding, automation and real-time applications.
OpenAI positions these models for high-volume environments where efficiency plays a key role. They support coding assistants, automation systems, and multimodal tools that need fast responses and stable performance without significantly increasing overall operational costs.
GPT-5.4 mini builds on GPT-5 mini with improved capabilities and faster performance. The model operates at more than twice the speed while enhancing results across coding, reasoning, tool use and multimodal understanding tasks.
In benchmark comparisons, GPT-5.4 mini scored 54.4% on SWE-Bench Pro, up from 45.7% earlier. It also reached 60% on Terminal-Bench 2.0 and achieved 88% on GPQA Diamond, the company said.
OpenAI says GPT-5.4 nano is built for simpler tasks like classification, ranking, and data extraction. It focuses on speed and lower costs while still delivering solid performance, making it suitable for large-scale and budget-sensitive deployments.
GPT-5.4 nano scored 52.4% on SWE-Bench Pro and 46.3% on Terminal-Bench 2.0. These results show it maintains stable performance while focusing on faster execution and lower costs for large-scale usage.
OpenAI says it uses a setup where bigger models like GPT-5.4 handle planning, while smaller ones like GPT-5.4 mini do tasks at the same time, helping improve speed and overall performance in complex workflows.
GPT-5.4 mini supports multimodal inputs and can understand screenshots for computer-based tasks. It also helps developers fix bugs, edit code, and work through large codebases more easily in real-world workflows.
GPT-5.4 mini is available via API, Codex and ChatGPT with a 400k context window, priced at $0.75 input and $4.50 output per million tokens. GPT-5.4 nano costs $0.20 input and $1.25 output via API.