OpenAI drops GPT-5.4 mini, nano

OpenAI is turning model families into a real operating model for agentic systems: one frontier model for planning, smaller models for execution, and nano for the boring glue work in between.

• –GPT-5.4 mini is positioned as more than a budget tier; OpenAI says it runs over 2x faster than GPT-5 mini while getting close to GPT-5.4 on several benchmarks.
• –The subagent framing matters more than the raw benchmark gain: this release is optimized for code search, targeted edits, screenshot interpretation, and other support tasks that benefit from low latency.
• –Nano looks deliberately narrow, which is exactly what many production pipelines need for classification, ranking, and entity extraction at scale.
• –Availability across API, Codex, and ChatGPT makes mini the obvious default workhorse for developers, while nano stays infrastructure-like rather than end-user visible.
• –The release reinforces a trend we’ve seen across AI tooling: the winning stack is increasingly a router, not a single monolithic model.