OpenAI is shrinking GPT-5.4 without scaling back its ambitions. The AI company has launched GPT-5.4 mini and GPT-5.4 nano, introducing faster, cheaper versions of its flagship model built for high-volume workloads where speed and cost matter most.
According to OpenAI, the new models inherit many of GPT-5.4’s strengths while targeting coding, subagents, multimodal tasks, and other jobs that require quick response times without the heavier price tag.
Built for different kinds of work
The split is fairly clear. GPT-5.4 mini is the more capable of the two, with OpenAI reporting gains in coding, reasoning, multimodal understanding, and tool use, while noting that it runs more than twice as fast as GPT-5 mini and comes close to the larger GPT-5.4 model on several evaluations.
GPT-5.4 nano, by contrast, is the leaner option built for jobs where speed and price carry more weight than breadth. The AI company calls it the smallest and cheapest version of GPT-5.4 and says it is meant for classification, data extraction, ranking, and coding subagents handling simpler supporting tasks, differentiating the AI model that takes on broader workloads from the one that keeps lighter tasks moving in the background.
From coding loops to computer use
GPT-5.4 mini and nano are especially well-suited to coding jobs that depend on fast iteration, including targeted edits, codebase navigation, front-end generation, and debugging loops.
The setup becomes more layered in Codex, where a larger model can handle planning, coordination, and final judgment while GPT-5.4 mini subagents take on narrower tasks in parallel, such as searching a codebase, reviewing large files, or processing supporting documents.
OpenAI highlights computer use as another lane for mini, saying it can interpret screenshots of dense interfaces to complete screen-based tasks quickly.
Where to get them and what they cost
The launch reaches users in very different ways depending on which model you are looking at.
GPT-5.4 mini gets the wider release across the API, Codex, and ChatGPT. In the API, it includes:
- Text and image inputs
- Tool use and function calling
- Web search and file search
- Computer use and skills
- A 400k context window
Its API pricing comes in at:
- $0.75 per 1M input tokens
- $4.50 per 1M output tokens
In Codex, mini is available in the app, CLI, IDE extension, and web, and uses 30% of the GPT-5.4 quota, putting simpler coding tasks at about a third of the cost. In ChatGPT, it is available to Free and Go users through the Thinking option in the plus menu, while other users get it as a rate-limit fallback for GPT-5.4 Thinking.
GPT-5.4 nano is more limited by design. It is API-only, with pricing set at:
- $0.20 per 1M input tokens
- $1.25 per 1M output tokens
The launch shows OpenAI placing more emphasis on where models fit in the stack, not just on how powerful the top model gets.
A new suit from Britannica turns up the pressure on OpenAI in the widening battle over copyrighted material and AI outputs.