OpenAI Launches GPT-5.4 Mini and Nano for Efficient Coding and Rapid Subagent Workflows

OpenAI expanded its model family on March 17, 2026, with the debut of GPT-5.4 mini and nano. These models bring the core strengths of the flagship GPT-5.4 architecture to faster, more efficient packages designed for high-volume workloads. By prioritizing low latency, these versions enable highly responsive experiences in coding assistants and computer-using systems.

The GPT-5.4 mini variant delivers a significant performance boost, operating more than twice as fast as the previous mini generation. In technical evaluations like SWE-Bench Pro, it achieved a 54.4% pass rate, which is remarkably close to the 57.7% score of the full flagship model. It also includes a 400,000 token context window for handling complex, data-heavy tasks.

For tasks where cost and speed are the primary drivers, GPT-5.4 nano offers an ultra-efficient solution priced at only $0.20 per million input tokens. This model is ideal for classification, ranking, and routing tasks that do not require full-scale reasoning effort. It allows for massive throughput in automated pipelines that handle data extraction at scale.

These models are specifically built for the emerging era of orchestrated AI development. Instead of using a single large model for every step, developers can use the flagship GPT-5.4 for planning while delegating execution to mini and nano subagents. This pattern is natively supported in the OpenAI Codex environment to improve overall iteration speeds. Availability begins immediately for API users and Codex subscribers, with the mini model also serving as a fallback for ChatGPT users. Enterprise teams will find these models available through the OpenAI API for building latency-sensitive multimodal applications. The aggressive pricing structure is specifically designed to encourage the deployment of large-scale agentic teams.