GPT-5.6 Luna — Specs, Pricing & Limits
Specs
| Status | Priced |
|---|---|
| Released (API changelog) | 2026-07-09 |
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | Feb 16, 2026 |
| Reasoning effort levels | none, low, medium (default), high, xhigh, max |
| Long-context rule | >272,000 input tokens: 2x input, 1.5x output (full request) |
| Cache writes | Billed at the cache-write rate instead of the input rate |
| ChatGPT credit rate (per 1M) | 5 in · 0.5 cached · 30 out |
Pricing per 1M tokens, every published service tier
| Standard — short context | in $0.20 · cached $0.02 · writes $0.25 · out $1.20 |
|---|---|
| Standard — long context | in $0.40 · cached $0.04 · writes $0.50 · out $1.80 |
| Batch — short context | in $0.10 · cached $0.01 · writes $0.125 · out $0.60 |
| Batch — long context | in $0.20 · cached $0.02 · writes $0.25 · out $0.90 |
| Flex — short context | in $0.10 · cached $0.01 · writes $0.125 · out $0.60 |
| Flex — long context | in $0.20 · cached $0.02 · writes $0.25 · out $0.90 |
| Fast — short context | in $0.40 · cached $0.04 · writes $0.50 · out $2.40 |
| Fast — long context | in $0.80 · cached $0.08 · writes $1.00 · out $3.60 |
| Cache writes, 12-hour lifetime | $0.50 Standard short · $1.00 long — 2.5× the tier's input rate, instead of 1.25× for the default 30-minute lifetime (Responses API) |
Rate limits by usage tier
| Free | Not supported |
|---|---|
| Build | 5,000 RPM · 2,000,000 TPM · 20,000,000 batch queue |
| Launch | 10,000 RPM · 10,000,000 TPM · 1,000,000,000 batch queue |
| Grow | 30,000 RPM · 180,000,000 TPM · 15,000,000,000 batch queue |
GPT-5.6 Luna is the cheapest model in the GPT-5.6 family, and the detail worth knowing before assuming "cheapest" means "most restricted" is that Luna sits in a different rate-limit group from Sol and Terra — one whose top-tier request and token ceilings and batch queues are higher, not lower.
That group is now shared with GPT-6 Luna, its successor, and the comparison between the two is unusually one-sided: GPT-6 Luna undercuts GPT-5.6 Luna on every published price cell, on the same context window and the same rate-limit table, with a newer knowledge cutoff. In Codex with ChatGPT sign-in, GPT-6 Luna is also the model OpenAI names for the Free and Go plans. GPT-5.6 Luna remains available during the GPT-6 rollout, and keeping it is reasonable for a high-volume pipeline you've already validated, where a model switch means re-running evaluations you'd rather not repeat.
Luna is built for volume — bulk classification, extraction, transformation, short structured outputs — where the rate-limit ceiling is more likely to bind than the per-token price. It is a poor fit for work that specifically needs Sol-class reasoning on a hard, multi-step task. Its effort levels span none to max, with medium as the default.
The long-context rule applies to Luna as it does to the rest of the family: crossing the threshold reprices the full request, not just the portion above the line.
The table above is computed directly from this site's facts module: pricing across every published service tier, the long-context bands, and Luna's own rate-limit table.
Verified 2026-10-28 against https://developers.openai.com/api/docs/models/gpt-5.6-luna.
Checked: https://developers.openai.com/api/docs/models/gpt-5.6-luna · https://developers.openai.com/api/docs/pricing