Migrating: GPT-5.4 Mini to GPT-5.6 Luna
Green marks the cheaper/newer side per row where a direction genuinely applies
| Context window | 400,000 tokens | 1,050,000 tokens |
|---|---|---|
| Knowledge cutoff | Aug 31, 2025 | Feb 16, 2026 |
| Standard input price | $0.75/MTok | $0.20/MTok |
| Standard cached input | $0.075/MTok | $0.02/MTok |
| Standard output price | $4.50/MTok | $1.20/MTok |
| Cache writes | No surcharge | Surcharged (write rate) |
| Reasoning effort levels | none (default), low, medium, high, xhigh | none, low, medium (default), high, xhigh, max |
| Long-context repricing rule | Not published | Published |
| Rate limit (Build tier) | 5,000 RPM · 2,000,000 TPM | 5,000 RPM · 2,000,000 TPM |
| API status | Available | Available |
| Codex with ChatGPT sign-in | Retired 2026-08-31 | No retirement announced |
A budget-tier move with a correction attached: this page originally said the price goes up and the rate-limit group changes. Neither is true. GPT-5.6 Luna is cheaper than GPT-5.4 Mini on every published price cell, and both models use the same Luna-group rate-limit table.
What actually moves
The context window grows from Mini's smaller window to the full GPT-5.6 window — a real capability gain. The knowledge cutoff moves forward by several months. Price falls on input, cached input and output at every tier both publish, Fast included. Effort gains a max level, and the default moves from none to medium: an unspecified request that did no reasoning on Mini will reason on Luna, which changes token use as well as behavior.
What's new to budget
Luna bills cache writes at its write rate, which sits above its own input rate; Mini adds no surcharge. Even so, Luna's write rate is below Mini's ordinary input rate, so caching costs less after the move, not more. Luna also publishes a long-context rule, where Mini publishes none — so a request whose input crosses the threshold now reprices in full, a rule that simply didn't exist on Mini's page.
What stays put
The rate-limit table, row for row, and the availability of a Fast tier.
The step after this one
In Codex with ChatGPT sign-in, GPT-5.4 Mini is already gone — it retired there on August 31, 2026 — and OpenAI's named replacement is GPT-6 Luna, not GPT-5.6 Luna. GPT-6 Luna runs on the same rate-limit table at a lower price still, so if you're migrating anyway, evaluate it alongside.
Verified 2026-10-28 against CodexHow facts module (src/data/facts/) — see /about/#accuracy.