OpenAI has released two smaller artificial intelligence models-GPT-5.4 mini and GPT-5.4 nano-expanding the GPT-5.4 family with options designed for faster responses and lower running costs.
OpenAI is positioning the pair for high-volume use cases where latency affects user experience, such as coding assistants, multimodal apps that work with images, and systems that operate software by interpreting screenshots.
GPT-5.4 mini sits between the flagship GPT-5.4 model and the earlier GPT-5 mini. OpenAI says it improves on GPT-5 mini across coding, reasoning, multimodal understanding, and tool use, while running at more than twice the speed.
OpenAI also says GPT-5.4 mini approaches GPT-5.4 performance on some external evaluations, including SWE-Bench Pro and OSWorld-Verified-benchmarks widely cited for software engineering tasks and computer-use agent performance.
GPT-5.4 nano is the smallest and lowest-priced model in the GPT-5.4 line. OpenAI describes it as a significant upgrade over GPT-5 nano and recommends it for classification, data extraction, ranking, and coding subagents that handle simpler supporting tasks.
Coding focus
Much of the announcement focused on software development workflows. OpenAI frames GPT-5.4 mini and nano as suited to fast-iteration tasks such as targeted code edits, navigating a codebase, generating front-end code, and running debugging loops.
Benchmark results, according to OpenAI, show GPT-5.4 mini consistently outperforming GPT-5 mini at similar latencies. The company also says it reaches pass rates close to GPT-5.4 while running much faster.
The emphasis reflects a broader trend in developer tooling: teams increasingly combine models of different sizes in one product, using larger models for planning and decision-making while offloading repetitive or narrow tasks to smaller, cheaper models.
Subagent pattern
OpenAI highlighted this approach in Codex, its coding product. In this setup, a larger model such as GPT-5.4 can handle planning and final judgement, while GPT-5.4 mini takes parallel subtasks such as searching a codebase, reviewing a large file, or processing supporting documents.
OpenAI says the pattern becomes more valuable as smaller models improve, and positions GPT-5.4 mini as its strongest "mini" model for systems built from multiple specialised agents.
Computer use
OpenAI also pointed to GPT-5.4 mini's multimodal features, saying it can interpret screenshots of dense user interfaces and complete computer-use tasks quickly. On OSWorld-Verified, the company says GPT-5.4 mini approaches GPT-5.4 and performs substantially better than GPT-5 mini.
API pricing
GPT-5.4 mini is available in the OpenAI API, Codex, and ChatGPT. In the API, it supports text and image inputs, as well as tool use, function calling, web search, file search, computer use, and skills. OpenAI says it has a 400k context window.
In the API, GPT-5.4 mini costs USD $0.75 per 1 million input tokens and USD $4.50 per 1 million output tokens.
GPT-5.4 nano is available only in the API, priced at USD $0.20 per 1 million input tokens and USD $1.25 per 1 million output tokens.
Codex and ChatGPT
In Codex, GPT-5.4 mini is available across the Codex app, command line interface, integrated development environment extension, and web offering. OpenAI says it uses 30% of the GPT-5.4 quota, allowing developers to run simpler coding tasks at roughly one-third of the cost compared with GPT-5.4 within Codex usage limits.
In ChatGPT, GPT-5.4 mini is available to Free and Go users via the "Thinking" feature in the + menu. For other users, it is available as a rate-limit fallback for GPT-5.4 Thinking.