GPT-5.4 Nano — Specs, Pricing & Limits

GPT-5.4 Nano was the cheapest model this site priced until GPT-6 Luna arrived. Luna now undercuts it on every Standard price cell — input, cached input and output — while offering a larger context window and a newer knowledge cutoff, so the case for Nano has narrowed to workloads already validated on it. And that case now has an end date: OpenAI deprecated Nano on the API on October 1, 2026 and shuts it down on April 1, 2027, naming GPT-6 Luna as the replacement.

No Fast mode is offered for this model at all, which is worth knowing if your mental model of "which models skip Fast" is that only the largest, most expensive ones do. Its rate-limit table is now exactly the Luna group's: the smaller entry row it used to have went with Tier 1 when OpenAI moved to Build, Launch and Grow.

No long-context repricing rule is published for this model, and the pricing page carries no long-context band for it — checked directly against both surfaces rather than assumed from the window size.

For genuinely simple, high-volume calls — classification, short extraction, lightweight formatting — the per-token arithmetic now favors GPT-6 Luna, and the remaining reasons to keep Nano until its shutdown are evaluation continuity and the none-by-default reasoning effort a pipeline may have been tuned around. Run the same task through both now rather than in the last weeks before the shutdown.

The table above is computed directly from this site's facts module: pricing across every published service tier and the rate-limit table.