GPT-5.6 Luna is a model in the OpenAI GPT-5.6 family focused on speed and cost efficiency, suitable for high-frequency Q&A, information extraction, text and image understanding, and everyday coding assistance. It is not merely a lightweight option for simple chatting; it also offers professional work and programming capabilities. On this platform, you can choose standard chat, Responses, or simplified conversation access based on application needs.
Clarify capacity, input and output, and invocation methods before choosing a model.
Family positioning is an inherent model characteristic, while interface formats and control fields are defined by this platform's endpoints. Optional controls must use configurations actually accepted by gpt-5.6-luna; tool calls only generate invocation requests, while permission checks, tool execution, and result return must still be completed by the application.
Learn what gpt-5.6-luna can bring to your work.
Luna's core trade-off is handling a large volume of everyday tasks with speed and cost efficiency, rather than pursuing the highest score on every difficult problem. It is well suited to organizing messy text into summaries, classifications, and to-do lists, and can also assist with professional work. When designing applications, let it handle routine requests with clear rules, and assign tasks requiring in-depth reasoning to higher-capability tiers.
Luna has substantial programming capabilities and is suitable for explaining code, reviewing localized changes, drafting tests, and analyzing errors. Providing relevant code, runtime behavior, and acceptance criteria is more helpful than simply saying “fix the issue.” It can serve as an everyday assistant in the development workflow, but for cross-module architecture or long-chain debugging, use test results to decide whether to upgrade the model.
Luna supports understanding images together with text questions, and can be used for explaining interface screenshots, organizing chart highlights, and answering questions about visual content. In standard conversations, images can be provided through image_url, along with the area of focus and desired answer format. Its delivery focus is text analysis rather than image generation; small text and dense charts should be cropped and enlarged before submission.
Start with specific tasks to find where the model can be effective.
Provide customer conversations, product rules, and fixed labels, and ask Luna to output issue types, key information, missing materials, and reply drafts. It is suitable for processing large volumes of similarly formatted records. When writing to business systems, a JSON structure can be specified and required fields validated by the program; operations such as refunds and authorizations should be handled by business rules and human approval.
Submit a code diff, related functions, and error logs, and have Luna organize the review results by issue location, impact scope, repair suggestions, and test points. The deliverable can be a review checklist or a patch draft for developers to continue validating. If an issue depends on substantial code that was not provided, add the relevant context first rather than asking the model to guess the entire project.
Provide a page screenshot together with the issue encountered by the user, and ask Luna to explain the visible state, possible operational misunderstandings, and next troubleshooting steps. It is suitable for creating help documentation, defect records, and interaction feedback. Prompts should distinguish between content directly observable in the image and causes that need to be inferred, resulting in text recommendations that support staff can easily use.
Choose based on task complexity, input materials, and expected results.
For frequent summarization, classification, localized code assistance, and routine text-and-image Q&A, you can start with Luna. Terra is the balanced tier of the same generation and is better suited to everyday professional work with higher complexity. The two are comparable in some comprehensive work evaluations, but the gap is more apparent in specialized tasks such as financial analysis. Therefore, do not judge substitutability based solely on being “of the same generation”; use real business samples for comparison.
When task rules are clear, the input scope is controllable, and results are easy to verify, Luna is the efficiency-first choice. If a task involves difficult reasoning, complex engineering coordination, or precise retrieval from large volumes of material, Sol’s flagship positioning is more worth considering. Luna should also not be understood as a comprehensive replacement for GPT-5.5: it has advantages in some coding and professional tasks, while other tasks still require specific comparison.
From a small-scale task to full integration.
Clarify objectives, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the full model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand output quality and capability scope.
Answers to common questions about using gpt-5.6-luna.
No. Luna, Terra, and Sol are different capability tiers in the same GPT-5.6 generation: Luna focuses on speed and cost efficiency, Terra focuses on balance, and Sol focuses on flagship capabilities. Use gpt-5.6-luna when calling it; you cannot turn it into Sol simply by changing the prompt, nor should you expect all three to perform the same on difficult tasks.
It is suitable for local code explanations, error analysis, change reviews, test drafts, and general implementation suggestions. It is best to provide the relevant code, runtime behavior, and acceptance criteria so the results can be verified. For cross-file dependencies, complex architectures, or repeated tool coordination, compare Terra and Sol, and make your choice based on whether actual tests pass.
Include both a text question and image_url image content in the message content of Chat Completions, and clearly specify the area to analyze and the desired output format. For example, ask it to list interface issues, explain chart trends, or organize information from a screenshot. Images should be clear and readable; small text and dense areas can be cropped before submission to reduce unnecessary visual distractions.
For existing messages-based conversation code, you can choose /openai/chat/completions; for input-based responsive interactions, you can choose /openai/responses. If you only need to submit questions and maintain a continuous conversation, you can use /aichat/conversations and continue the conversation through stateful and the returned id; the input organization methods for different endpoints should not be mixed.
You can ask Luna to generate JSON with specified fields for information extraction, classification, and ticket organization. Standard conversation endpoints define JSON object and JSON Schema format control fields; when using these controls, follow the configurations actually accepted by gpt-5.6-luna. The prompt should clearly specify field meanings, missing-value handling, and allowed labels. After receiving the result, the application still needs to parse the JSON, validate its structure and business rules, and handle refusals or truncated output; correct formatting does not mean the content is accurate, nor does it mean related actions have already been performed.
Model information · Updated: 2026-10-01. For calling parameters and billing rules, see the API and pricing sections.
Start with a clear goal, and use real results to determine whether it suits your work.
Input
$0.04per 1M tokens
Official $0.20
Output
$0.24per 1M tokens
Official $1.20
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs