Introducing GPT-5.4 mini and nano

OpenAI has unveiled GPT-5.4 mini and GPT-5.4 nano, expanding its model family with two highly efficient, small language models (SLMs) built for developers. The flagship of the pair, GPT-5.4 mini, is designed as a direct successor to GPT-4o mini, offering significantly improved performance at a lower cost. It processes information twice as fast as its predecessor and is priced aggressively at $0.15 per million input tokens and $0.60 per million output tokens. This makes it one of the most cost-effective options in OpenAI's portfolio for high-volume tasks like content moderation, classification, and simple customer support automation.

The even smaller GPT-5.4 nano model represents a push toward on-device and edge computing. While full specs are still emerging, it is engineered for scenarios where latency, cost, and data privacy are paramount, such as in mobile apps or IoT devices. Both models inherit advanced reasoning capabilities from the GPT-5.4 architecture, meaning they can handle complex logic and follow intricate instructions despite their compact size. This release signals OpenAI's strategy to capture the entire developer stack, from massive cloud-based models to tiny, specialized agents, directly competing with other efficient models like Google's Gemma 2 and Meta's Llama 3.1.

Early benchmarks indicate that GPT-5.4 mini not only beats its predecessor on speed but also shows marked improvements in coding and reasoning tasks, challenging the notion that smaller models must sacrifice capability. The launch provides a clear, scalable path for businesses to integrate AI into more products without ballooning infrastructure costs. By offering a 90% cost reduction compared to using larger models for similar tasks, OpenAI is lowering the barrier to entry and enabling a new wave of AI-powered features in everyday software.