What is GPT-5.6 Luna?
GPT-5.6 Luna is the cost-efficient model in OpenAI's GPT-5.6 family. It is designed for high-volume workloads and roughly corresponds to the nano tier in earlier GPT families. The model accepts text and images, produces text, supports reasoning effort controls and works with the Responses and Chat Completions APIs.
Luna is not simply a short-context utility model. OpenAI documents a 1,050,000-token context window, up to 922,000 input tokens and 128,000 output tokens. That combination makes it suitable for large document sets and repositories, but only when retrieval quality, latency and long-context pricing are measured carefully.
QUICK VERDICT: GPT-5.6 Luna is OpenAI's lowest-cost GPT-5.6 tier and a strong routing candidate for well-defined, high-volume work. Its million-token context and broad Responses API tool support are unusually capable at this price, but difficult reasoning and consequential decisions still need escalation to a stronger model or a human reviewer.
Field | Verified value |
|---|---|
Provider | OpenAI |
General availability | July 2026 |
License | Proprietary |
API model ID | gpt-5.6-luna |
Context window | 1,050,000 tokens |
Maximum input | 922,000 tokens |
Maximum output | 128,000 tokens |
Modalities | Text and image input; text output |
Knowledge cutoff | February 16, 2026 |
Reasoning effort | None, low, medium, high, xhigh and max; medium is default |
Current API price | $0.20 input, $0.02 cached input and $1.20 output per million tokens |
Last verified | October 22, 2026 |
What changed from GPT-5.4 nano
Compared with GPT-5.4 nano, Luna combines a much larger context window with the newer GPT-5.6 reasoning and tool stack. The practical improvement is broader task coverage at a low token rate, not a promise that it can replace a flagship model on every difficult workflow.
Area | GPT-5.6 Luna change |
|---|---|
Model position | Current low-cost tier in the GPT-5.6 family |
Context | 1.05 million tokens with a documented 922,000-token input limit |
Reasoning | Six effort levels from none through max |
Tools | Web search, file search, code execution, hosted shell, computer use, MCP and more through Responses |
Structured work | Structured outputs and function calling are supported |
Price | Current standard rates are $0.20 input and $1.20 output per million tokens |
A newer model is not automatically the right migration target. Compare GPT-5.6 Luna with GPT-5.4 nano using the same prompts, tools, source material, effort controls and acceptance rubric. Record quality, latency, retries, review time and total cost per accepted result.
Capabilities and best use cases
Luna supports streaming, structured outputs, function calling, prompt caching, image input, file search and web search. Through the Responses API it can also use code execution, hosted shell, apply-patch, skills, computer use, MCP connections and tool search. The safest deployments give it narrowly scoped tools and observable escalation criteria.
Capability | Practical use |
|---|---|
Classification and routing | Label requests, choose workflows and escalate uncertain cases |
Extraction | Convert text, documents and images into validated structured fields |
Support automation | Draft routine responses and gather information through approved tools |
Coding subtasks | Focused transformations, tests, patches and repository assistance |
Agent subtasks | Cheap parallel workers for bounded steps inside a larger workflow |
Long inputs | Large document sets or repositories when retrieval and evidence are tested |
The strongest documented fit is cost-sensitive classification, extraction, routing, support automation and focused agent subtasks. Reliability still depends on retrieval, tool permissions, prompt design, output validation and escalation policy. Test the complete system instead of evaluating the model name in isolation.
Benchmark evidence
OpenAI reports that Luna improves the price-performance tradeoff across coding and professional work, including comparisons with older, more expensive models. Those launch results are useful for choosing candidates, but they use OpenAI's harnesses, tools and effort settings. They should not be treated as a substitute for workload-specific testing.
Evaluation area | Reported evidence | How to interpret it |
|---|---|---|
Coding agents | Provider reports stronger results than older nano-tier models | Validate with the same repository, tools and acceptance tests |
Knowledge work | Provider reports competitive performance at lower estimated cost | Artifact quality and review time matter more than token price alone |
Computer use | Supported through Responses API tooling | Measure recovery, wrong clicks and approval behavior |
Long context | 1.05M documented context window | Capacity does not guarantee retrieval accuracy across the full prompt |
Reasoning effort | None through max supported | Higher effort can increase latency and output cost |
Benchmark results depend on the model version, effort, sampling, tools, scaffolding and dataset release. Do not combine scores from different harnesses into a synthetic ranking. Use public evidence to choose candidates, then test those candidates on frozen tasks from the intended workflow.
Pricing and access
GPT-5.6 Luna is available through the OpenAI API. Responses is the preferred endpoint when an application needs the complete tool surface; Chat Completions is supported for conventional chat-style integration. Batch is supported, while Realtime, fine-tuning and the legacy Assistants endpoint are not.
Prompts above 272,000 input tokens are billed at twice the standard input rate and 1.5 times the standard output rate for the full request. Cache writes cost 1.25 times uncached input. Large prompts therefore need explicit budgets and retrieval tests rather than automatic use of the full context window.
Usage or access item | Current value |
|---|---|
Standard input | $0.20 per million tokens |
Cached input | $0.02 per million tokens |
Cache writes | $0.25 per million tokens |
Standard output | $1.20 per million tokens |
Prompts above 272K | 2x input and 1.5x output for the full request |
Batch | Supported; verify current processing discount in OpenAI pricing |
Token price is only part of deployment cost. Include prompt construction, cached-input behavior, tool calls, retries, wall time, failed-task recovery and human review. A cheaper model can be more expensive operationally when it needs repeated attempts or produces artifacts that require substantial correction.
Limitations and deployment risks
- Luna is optimized for cost-sensitive work, not the hardest reasoning or judgment-heavy tasks.
- A million-token context window can create large latency and cost without improving answer quality.
- Prompts above 272,000 tokens trigger higher rates across the full request.
- The model does not support fine-tuning, Realtime or the legacy Assistants endpoint.
- Tool access requires least-privilege permissions, confirmation gates and recoverable operations.
- Low token price can be offset by retries, review, tool calls and silent task failures.
- Provider benchmark claims need reproduction on the actual deployment workflow.
High-stakes legal, medical, financial, scientific and security work needs qualified review. Log the exact model version and settings, separate untrusted content from system instructions, minimize tool permissions and require confirmation before irreversible or externally visible actions.
How to evaluate GPT-5.6 Luna
Build a frozen evaluation set of 20 to 50 real tasks. Include routine work, difficult edge cases, adversarial inputs, long-context examples and cases where the correct behavior is to stop or escalate. Compare GPT-5.6 Luna with GPT-5.4 nano and at least one neighboring model under equivalent conditions.
Test area | What to record |
|---|---|
Task completion | Pass or fail against a written acceptance rubric |
Reliability | Repeated-run success rate, variance and silent failures |
Factuality | Unsupported claims, source use, quotations and citation accuracy |
Tool use | Wrong calls, retries, recovery and permission-boundary failures |
Long context | Retrieval accuracy, instruction retention and cost at realistic lengths |
Efficiency | Wall time, input, output, cached tokens, tool charges and review time |
Safety | Prompt injection, sensitive-data handling and irreversible-action controls |
Choose the least expensive configuration that clears the acceptance threshold with a safety margin. Re-run the suite when the model, system prompt, effort, retrieval layer, tool definitions or approval policy changes.
Who should use it?
Situation | Recommendation |
|---|---|
Strong fit | cost-sensitive classification, extraction, routing, support automation and focused agent subtasks |
Pilot first | Long-running agents, large contexts, computer use and workflows with several tools |
Escalate | Ambiguous or consequential work that does not reliably clear the evaluation threshold |
Avoid unsupervised use | Irreversible actions, sensitive data or high-stakes decisions without monitoring and approval |
A migration should be driven by measured outcomes. Keep GPT-5.4 nano available during the pilot, record where each model succeeds or fails, and use routing when different task classes have different quality and cost requirements.
Frequently asked questions
Is GPT-5.6 Luna open source?
No. GPT-5.6 Luna is a proprietary model controlled by OpenAI. Access, lifecycle and serving behavior depend on the provider and supported distribution platforms.
How much does GPT-5.6 Luna cost?
$0.20 per million tokens. Review the full table because caching, prompt length, processing mode, platform and output volume can materially change the total.
What is GPT-5.6 Luna best used for?
Its strongest documented fit is cost-sensitive classification, extraction, routing, support automation and focused agent subtasks. Start with a supervised pilot and retain human sign-off for consequential work.
Should I migrate from GPT-5.4 nano?
Only after a side-by-side evaluation. Measure accepted-task quality, total cost, latency, output style, tool reliability and migration engineering. Newer does not always mean better for a specific operational workflow.
How should benchmark evidence for GPT-5.6 Luna be interpreted?
Use provider results to identify promising workloads, then reproduce the comparison with the exact configuration, tools and acceptance criteria that matter to the deployment. Do not infer a site ranking from benchmark claims gathered under a different harness.
Related model guides
Browse the Best AI Models directory.
Compare with the GPT 5.6 Sol guide.
Compare with the GPT 5.5 guide.
Compare with the GPT-6 Luna guide.
Official sources and update policy
OpenAI guide to the current GPT model family.
GPT-5.6 Luna API model documentation.
Checked October 22, 2026. We update this guide when OpenAI changes the specification, pricing, access, safety documentation or lifecycle. Vendor benchmark claims are attributed and are not presented as independent testing. No vendor payment or affiliate relationship determined inclusion.