OpenAI Launches GPT-5.4 Mini and Nano – Faster AI for Everyone

OpenAI has released two new models – GPT-5.4 mini and GPT-5.4 nano – built for high-speed, cost-efficient AI workloads. The launch on March 17, 2026 represents a strategic shift: flagship-level performance in lighter form factors.

What's New?

GPT-5.4 mini delivers performance close to the flagship GPT-5.4 model at a fraction of the cost and latency. It scores 88.01% on the GPQA Diamond benchmark (advanced expertise measurement), just behind the flagship's 93%. The model excels especially at coding, reasoning, and multimodal tasks.

GPT-5.4 nano is the smallest and fastest variant, optimized for data-intensive applications and real-time integrations. It's ideal for systems requiring instant responses – like interpreting screenshots or processing large documents in real time.

Availability

GPT-5.4 mini is now available to all ChatGPT users, including Free tier, via the "Thinking" option. Nano is developer-targeted via API for high-volume workloads.

The Subagent Strategy

OpenAI highlights a new usage model: large models act as "senior managers" delegating narrow, repetitive tasks to mini and nano as "subagents." Notion already reports that GPT-5.4 mini handles focused tasks with impressive accuracy.

CIO Perspective

For enterprises this means: you can build agentic pipelines where advanced reasoning combines with cheap, fast execution. The cost of AI automation drops dramatically. With GPT-5.4 nano, high-volume operations like document processing and code review can scale without proportional budget increases.