OpenAI has released GPT-5.4 mini and GPT-5.4 nano, calling them its “most capable small models yet,” with a clear focus on speed, efficiency, and strong performance for everyday AI tasks, while also making these models easier to use across ChatGPT, APIs, and developer tools.
According to OpenAI, these new models target high-volume workloads where fast responses and lower costs matter, while still improving results across coding, reasoning, and multimodal tasks.
GPT-5.4 mini builds on the earlier GPT-5 mini with clear upgrades across coding, reasoning, multimodal understanding, and tool use, and it runs “more than 2x faster,” which makes it useful for real-time workflows and frequent queries. OpenAI also says it “approaches the performance” of the larger GPT-5.4 model on benchmarks like SWE-Bench Pro and OSWorld-Verified, which shows how much smaller models have improved.
You can access GPT-5.4 mini today in ChatGPT, Codex, and OpenAI’s API, and Free and Go users will find it under the “Thinking” option, while other users will see it as a fallback for GPT-5.4 Thinking.
GPT-5.4 nano targets low-cost and simple tasks
GPT-5.4 nano is the smallest and cheapest version in the lineup, and OpenAI recommends it for tasks like classification, data extraction, ranking, and lightweight coding support, where “speed and cost matter most.” It improves on GPT-5 nano and fits well for backend processes and simple agent workflows.
OpenAI continues to expand its model lineup, with GPT-5.4 Thinking adding six improvements earlier this month, GPT-5.3 Instant launching with a focus on cleaner responses, and the Codex Mac app arriving earlier this year, showing a steady push toward faster and more practical AI tools.
