GPT-5.4 lands with Pro, Thinking, mini, nano

OpenAI released GPT-5.4 on March 5, 2026, then added mini and nano on March 17.

OpenAI launched GPT-5.4 on March 5, 2026, starting with GPT-5.4 Thinking and GPT-5.4 Pro. Neither tier was open to free users at launch.

What changed

On March 17, OpenAI added GPT-5.4 mini and GPT-5.4 nano. GPT-5.4 mini was made available to free-tier users, while GPT-5.4 nano shipped only through the OpenAI API.

OpenAI said the new model family was tuned for professional workflows. Reported changes included built-in computer use, stronger deep research, and a 33% drop in factual errors versus GPT-5.2.

• GPT-5.4 Thinking and Pro launched first.
• GPT-5.4 mini reached free users later.
• GPT-5.4 nano stayed API-only.
• API pricing for mini and nano was four times GPT-5 equivalents.

Benchmark results also pointed to better desktop control. In OSWorld-Verified, GPT-5.4 scored 75%, above GPT-5.2 at 47.3% and slightly above the cited human average of 72.4%.

Why it matters

For developers building agents, the release signals a push toward models that can do more than answer text prompts. Built-in computer use and stronger research features make GPT-5.4 more relevant for workflows that touch spreadsheets, browser tasks, and desktop apps.

The pricing split matters too. Free access via GPT-5.4 mini broadens reach, but the 4x API premium on mini and nano raises the cost of using OpenAI’s newest smaller models in production.

The core question now is whether GPT-5.4’s gains in accuracy and computer use are enough to justify the higher API bill for teams that need speed, scale, and agent-like behavior.