OpenAI Launches GPT-5.4 mini and nano

OpenAI has just launched GPT-5.4 mini and GPT-5.4 nano, designed to bring flagship-level capabilities to high-volume, low-latency applications. GPT-5.4 mini, in particular, delivers a dramatic leap in performance, running more than twice as fast as its predecessor.

While the primary GPT-5.4 “Thinking” model remains the powerhouse for deep reasoning, the mini and nano variants are built to be the “workhorses” of the AI world. One of the most impressive aspects of the mini model is how closely it mirrors the intelligence of the full-scale GPT-5.4.

In key benchmarks like SWE-Bench Pro, which tests real-world software engineering capabilities, GPT-5.4 mini scored 54.4%, rivaling the 57.7% achieved by the much larger flagship model. This makes it an ideal choice for “vibe coding” and responsive coding assistants where speed is as critical as accuracy.

The GPT-5.4 series introduces a unified approach to AI, integrating advanced coding, reasoning, and tool use into single systems.

• GPT-5.4 mini: Available today for ChatGPT Free and Go users via the “Thinking” menu, it supports text and image inputs, web search, and even the ability to interact with computer software directly through screenshots and mouse clicks.
• GPT-5.4 nano: Aimed at developers, this is the smallest and most affordable version of GPT-5.4. It is specifically recommended for high-volume tasks like classification, data extraction, and acting as a “subagent” that handles simple supporting tasks within a larger AI workflow.

OpenAI’s goal is to allow developers to mix and match models, using “Thinking” versions for complex planning and delegating simpler steps to the cheaper mini or nano subagents. This strategy effectively lowers the cost of operating complex, multi-step workflows at scale.