Last updated: July 7, 2026
GPT-5.6 Luna is OpenAI’s fast, affordable model in the GPT-5.6 family. It is the model to watch if your main question is not “What is the smartest GPT-5.6 model?” but “How do we run GPT-5.6-style workflows at high volume without sending everything to the flagship tier?”
Quick answer: GPT-5.6 Luna’s preview model ID is gpt-5.6-luna. OpenAI lists preview pricing at $1 per 1M input tokens and $6 per 1M output tokens. Luna is intended for speed and cost efficiency. During the preview, GPT-5.6 models are available only to selected trusted partners and organizations through the OpenAI API and/or Codex, and OpenAI says GPT-5.6 is not available in ChatGPT during the preview.
Verified July 7, 2026
- OpenAI launch post: Previewing GPT-5.6 Sol
- OpenAI Help Center: A preview of GPT-5.6 Sol, Terra, and Luna
- Safety details: GPT-5.6 Preview System Card
- Model catalog status: OpenAI API models page
GPT-5.6 Luna quick facts
| Detail | GPT-5.6 Luna |
|---|---|
| Model family | GPT-5.6 |
| Model ID | gpt-5.6-luna |
| Role | Fastest and most cost-efficient GPT-5.6 model |
| Preview input price | $1.00 / 1M tokens |
| Preview output price | $6.00 / 1M tokens |
| Cache write | $1.25 / 1M tokens |
| Cached input read | $0.10 / 1M tokens |
| Preview access | Selected API/Codex partners and organizations |
| ChatGPT access | Not included during preview |
| Best for | High-volume tasks, extraction, tagging, first drafts, routing |
| Escalate to Terra when | The task needs more nuance or professional-quality reasoning |
| Escalate to Sol when | The task is hard, high-value, or deeply agentic |
What is GPT-5.6 Luna?
GPT-5.6 Luna is the low-cost tier of the GPT-5.6 family. OpenAI describes Luna as the fastest and most cost-efficient model in the family. That makes it useful for a different class of work than Sol.
A Sol-first workflow asks: “How do we get the best answer?”
A Luna-first workflow asks: “How do we solve many simple tasks cheaply, then escalate only the hard ones?”
That is the right way to evaluate Luna. It is not supposed to beat Sol at the hardest tasks. It is supposed to make GPT-5.6 economics work for high-volume systems.
GPT-5.6 Luna pricing
OpenAI lists Luna at:
| Token type | Price per 1M tokens |
|---|---|
| Input | $1.00 |
| Output | $6.00 |
| Cache write | $1.25 |
| Cached input read | $0.10 |
Luna is 5x cheaper than Sol on input tokens and 5x cheaper on output tokens. Compared with Terra, Luna is 2.5x cheaper on input and output tokens.
That price difference matters when you run:
- millions of classification requests;
- customer support triage;
- lead scoring;
- content tagging;
- short extraction jobs;
- first-pass summaries;
- routing decisions before a larger model.
Luna vs Terra vs Sol
| Model | Main role | Best use |
|---|---|---|
| GPT-5.6 Sol | Flagship | Hard reasoning, agentic coding, research, security, high-value tasks |
| GPT-5.6 Terra | Balanced | Everyday professional work, analysis, coding help, documents |
| GPT-5.6 Luna | Fast/low-cost | High-volume tasks, extraction, tagging, cheap drafts, routing |
Luna should usually be the first model in a cost-saving router, not the only model in the system.
A simple routing pattern:
- Ask Luna to classify the task difficulty.
- Let Luna handle easy tasks directly.
- Escalate medium tasks to Terra.
- Escalate hard or high-risk tasks to Sol.
- Use human review for regulated, sensitive, or customer-facing outputs.
Best GPT-5.6 Luna use cases
Luna is a strong candidate for:
- intent classification;
- sentiment classification;
- lead scoring;
- spam or abuse triage;
- short summaries;
- metadata generation;
- document routing;
- extracting fields from structured text;
- tagging support tickets;
- first-draft email replies;
- generating title options;
- transforming content formats;
- quick QA checks before a larger model runs.
The common pattern is that the task is cheap to verify. If a simple validator or human reviewer can catch mistakes quickly, Luna’s low cost becomes valuable.
When not to use Luna
Do not use Luna as the only model when:
- the task is high-stakes;
- the output is hard to verify;
- the prompt involves legal, financial, medical, cyber, or scientific decisions;
- the task needs long planning or tool orchestration;
- the user expects polished final work;
- a bad answer can create expensive downstream damage.
In those cases, Luna may still be useful as a classifier or preprocessor, but the final answer should likely come from Terra, Sol, or a human-reviewed workflow.
Luna and prompt caching
Luna’s cached-input pricing is especially interesting for high-volume workflows. OpenAI’s preview Help Center page says GPT-5.6 and later models support more predictable prompt caching with explicit cache breakpoints and a 30-minute minimum cache life.
For Luna, cached reads are listed at $0.10 per 1M tokens. That makes repeated shared context much cheaper.
Good candidates for cacheable context include:
- a support policy;
- a product catalog summary;
- a classification rubric;
- a brand style guide;
- a schema definition;
- a workflow instruction block;
- a safety or escalation policy.
The best Luna systems will not send a giant fresh prompt every time. They will cache stable context, keep outputs short, and escalate only when needed.
Should you test GPT-5.6 Luna?
Test Luna if your team has preview access and your workload has enough volume for price to matter.
- Classification
- Extraction
- Ticket routing
- Cheap first drafts
- High-volume metadata
- Customer-facing responses
- Complex summaries
- Security-sensitive triage
- Hard reasoning
- Long autonomous workflows
- Medical, legal, or financial decisions without review
Bottom line
GPT-5.6 Luna is the cost-efficiency play in the GPT-5.6 family. It should not replace Sol for hard problems or Terra for nuanced professional work. But if you have thousands or millions of small tasks, Luna may be the model that makes the whole GPT-5.6 stack economically practical.
Use Luna for easy-to-check volume. Use Terra for the middle. Use Sol for the hard tail.
FAQ
What is GPT-5.6 Luna?
What is the GPT-5.6 Luna model ID?
gpt-5.6-luna.