OpenAI has unveiled GPT-5.4 mini and GPT-5.4 nano, describing them as its most capable small models so far. The company said the new releases combine many of the strengths of GPT-5.4 with an architecture built for speed, efficiency and heavy workloads.
According to the announcement, GPT-5.4 mini is aimed at coding, reasoning, multimodal analysis and tool use. OpenAI says it outperforms GPT-5 mini in these areas while running more than twice as fast. In benchmark results, GPT-5.4 mini reached 53.40% on SWE-bench Pro, 59.30% on Terminal-Bench 2.0 and 70.60% on OSWorld-Verified, bringing it closer to the full GPT-5.4 model in several tests.
The company said the mini version is designed for low-latency AI assistants, lightweight subagents, interface control systems that interpret screenshots, and real-time multimodal applications. OpenAI argues that in such cases, the best model is not always the largest one, but the one that responds immediately and works reliably with tools.
GPT-5.4 nano is presented as the smallest and fastest version in the lineup. It is intended for tasks where speed and cost matter most, including classification, data extraction, ranking and simple assistant workflows. OpenAI also shared early benchmark figures for nano, including 52.39% on SWE-bench Pro and 82.83% on GPQA Diamond, and said customers are already testing both models in production settings.
Read “Zamin” on Telegram!
