OpenAI Shrinks GPT-5.4 Into Mini and Nano Models Built for Speed

OpenAI Rounds Out GPT-5.4 Family With Mini and Nano

OpenAI has launched GPT-5.4 mini and nano, two smaller, faster versions of its flagship model released just two weeks ago. The new models are designed for high-volume workloads where speed and cost efficiency matter most.

Performance Leaps

GPT-5.4 mini delivers major improvements over its predecessor across coding, reasoning, multimodal understanding, and tool use, all while running more than twice as fast. On the SWE-Bench Pro coding benchmark, the mini model scores over fifty-four percent, approaching the full GPT-5.4's nearly fifty-eight percent. Its biggest jump comes in computer use capabilities, where it reaches over seventy-two percent on the OSWorld benchmark, a massive leap from the previous mini model's forty-two percent.

GPT-5.4 nano, the smallest in the lineup, targets tasks where latency and cost are paramount, including classification, data extraction, and ranking.

Higher Capability, Higher Cost

The improvements come at a price. GPT-5.4 mini costs three times more per input token than its predecessor, while nano's input pricing has quadrupled. Both models support a four hundred thousand token context window.

The Subagent Era

OpenAI is positioning these models for a new architectural pattern where a powerful model like GPT-5.4 handles planning and coordination, while delegating simpler parallel subtasks to mini or nano. In OpenAI's Codex platform, the mini model consumes only thirty percent of the full model's quota, cutting costs for routine coding tasks to roughly a third.

Availability

GPT-5.4 mini is available to free and paid ChatGPT users, in Codex, and through the developer API. Nano remains API-only for now.