Migrating: GPT-5.4 Mini to GPT-6 Luna

GPT-5.4 Mini left Codex for ChatGPT sign-in on August 31, 2026, and GPT-6 Luna is the replacement OpenAI names wherever your plan and client offer it. On the API, GPT-5.4 Mini is still available, so there this is an upgrade decision rather than a forced move — and the table above makes it an easy one.

What gets better

Nearly everything. GPT-6 Luna is cheaper on input, cached input and output, has a far larger context window, and its knowledge cutoff is the newest of any model on this site, many months after GPT-5.4 Mini's. It publishes the long-context repricing rule, which GPT-5.4 Mini's page doesn't state. It also adds max to the effort range.

What stays the same

Both sit in the same high-throughput Luna rate-limit group, so request and token ceilings at every usage tier carry over, and both offer Standard, Batch, Flex and Fast.

Settings to check

The default reasoning effort differs. GPT-5.4 Mini defaults to none; GPT-6 Luna defaults to medium. A request that never set effort explicitly will start reasoning after the switch — slower and with more output tokens than before. Set none explicitly if you want the old behavior; GPT-6 Luna accepts it.

Caching also changes shape. GPT-5.4 Mini has no separate cache-write rate, so written tokens bill at its input rate; GPT-6 Luna bills writes at its own rate, which is still well below GPT-5.4 Mini's input rate. On the API, review the newer cache-lifetime setting as well.

Enterprise and Edu

An administrator must enable GPT-6 Luna before members can select it in Codex. Until then, a config that names gpt-6-luna won't work for those users, however correct it is.