GPT-5.6 Luna vs GPT-5.4 Nano

These were the two cheapest models on this site until GPT-6 Luna arrived, and the comparison between them was misstated here at launch: this page called Nano the rock-bottom option with Luna a step up in price. On the published table they cost the same on input and cached input, and Luna is the cheaper of the two on output.

Window and recency favor Luna

Luna's context window is far larger than Nano's, and its knowledge cutoff is newer. A workload sized against Nano's window out of necessity loses that constraint by moving.

Rate limits are identical

Both use the Luna-group table at every paid tier. Nano's one smaller row was Tier 1's, which went away when OpenAI replaced Tier 1 to Tier 5 on October 6, 2026.

Caching and effort split them

Luna, as a GPT-5.6 model, bills cache writes at its write rate, which is above its input rate; Nano adds no write surcharge, so a written token costs Nano's ordinary input rate. That makes a write-heavy, read-light caching pattern marginally cheaper on Nano — the one bucket where it wins. Defaults differ too: Nano defaults to none effort, Luna to medium, and only Luna accepts max.

Fast mode

Luna offers Fast. Nano doesn't offer it at all.

How to decide

Luna matches or beats Nano on everything except cache writes, with a larger window and newer knowledge — and Nano is deprecated on the API, shutting down on April 1, 2027. Then check GPT-6 Luna, OpenAI's named replacement for Nano, which undercuts both on every Standard price cell.