Source reviewedProvider documentation

GPT-6 Luna

Q: Can GPT-6 Luna run without reasoning?

Yes, Luna supports reasoning.effort set to none, unlike GPT-6 Astra and GPT-6.1 Sol.

SOURCE-REVIEWED Checked against provider documentation · Not benchmarked

OPENAI · SPECS CHECKED 2026-10-15

GPT-6 Luna is the model OpenAI calls its "most efficient model for focused, high-volume tasks".

Compare models →View sources ↓

MODEL RECORD

Specifications & pricing.

CHECKED 2026-10-15

FIELD: DOCUMENTED VALUE; SOURCE
API identifier: gpt-6-luna (default and only listed snapshot); OpenAI model page
Input / output modalities: Text and image in; text out; OpenAI model page
Context window: 1,050,000 tokens (maximum input 922,000 tokens); OpenAI model page
Max output: 128,000 tokens; OpenAI model page
Knowledge cutoff: 18 May 2026; OpenAI model page
Reasoning controls: reasoning.effort: none, low, medium (default), high, xhigh, max; OpenAI model page
Input price: USD 0.10 per 1M tokens (list price as published on 15 October 2026); OpenAI model page
Cached input price: USD 0.01 per 1M tokens (10% of uncached input); OpenAI model page
Cache writes: USD 0.125 per 1M tokens (1.25x uncached input); OpenAI model page
Output price: USD 0.50 per 1M tokens; OpenAI model page
Batch / Flex / Fast: Batch and Flex at 50% of Standard; Fast at 2x; regional processing adds 10%; OpenAI model page
Endpoints: Responses, Chat Completions, Batch supported; Realtime, Assistants, fine-tuning, embeddings, audio and image endpoints not supported; OpenAI model page
Notable tools (Responses API): web_search, file_search, image_generation, code_interpreter, hosted_shell, apply_patch, skills, computer_use, mcp, tool_search; OpenAI model page

When to use it

What OpenAI says. The Using GPT-6 guide labels Luna "fastest and most cost-effective", with strong performance for focused, high-volume tasks. The model selection guide calls it the most cost-efficient model and maps two settings: Luna at low effort for fine-grained edits, well-scoped problem solving and simple data extraction; Luna at extra high effort for finding current context across multiple apps, prioritizing work and solving problems with clear constraints. The same guide recommends keeping "the lightest setting that meets your quality bar" after testing on the same inputs.

Our editorial judgment. Luna is the default worth trying for any pipeline that runs thousands of calls a day: tagging support tickets, pulling fields from documents, rewriting short copy, first-pass triage before a bigger model. Its large context window also means you can feed it long documents for narrow extraction jobs, though the long-prompt surcharge still applies above 272K tokens. Escalate to Sol when a task needs judgment across conflicting sources. If you are choosing a consumer assistant rather than an API model, our AI chatbots guide covers that.

Limits and costs to plan for

Long-prompt surcharge. Prompts above 272K input tokens are billed at 2x input and cache rates and 1.5x output for the full request: USD 0.20 input and USD 0.75 output per 1M tokens at list price.
Function calling in Chat Completions. Chat Completions supports function calling only with reasoning_effort set to none. Use the Responses API for built-in tools and for reasoning with tools.
Unsupported parameters. When reasoning effort is not none, remove temperature, top_p and top_logprobs, per the migration guidance.
Data residency. EU data residency is available with Standard, Fast, Flex and Batch processing. Regional processing adds a 10% premium where available.
Rate limits by tier. 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. These are the highest listed limits of the three GPT-6 models.
No audio, video or fine-tuning endpoints. The model page marks Realtime, audio, image generation endpoints and fine-tuning as not supported.

How it compares within the family

All three GPT-6 models share a 1,050,000 token context window, 128,000 max output, text and image input and the same Responses API tool list.

Model	Input / output (USD per 1M tokens)	Cached input	Reasoning `none`	Grow tier TPM
GPT-6 Astra	10.00 / 50.00	1.00	No	40,000,000
GPT-6.1 Sol	2.00 / 10.00	0.10	No	40,000,000
GPT-6 Luna	0.10 / 0.50	0.01	Yes	180,000,000

Luna's per-token price is one twentieth of Sol's and one hundredth of Astra's. It also has the latest knowledge cutoff of the three (18 May 2026 against 30 April 2026). Outside OpenAI, Grok 4.7 lists USD 2.00 input and USD 6.00 output per 1M tokens, so Luna is far cheaper per token, though the two are positioned for different work.

How to evaluate it on your work

Sample at least 100 real inputs from the high-volume job you want to move, with expected outputs labelled by a person.
Run Luna at none, low and medium effort, plus Sol at medium as a reference, using identical prompts and schemas.
Measure accuracy, latency and tokens per call for each setting, and compute cost per 1,000 correct results.
Check failure cases by hand to see whether errors cluster on a type of input you could route to a bigger model.
Pick the lowest effort that meets your accuracy bar, then test Batch or Flex for jobs that do not need an immediate response, since both are priced at 50% of Standard.

Method and limits

We read the GPT-6 Luna model page, the Using GPT-6 guide and the model selection guide on 15 October 2026. We have not run GPT-6 Luna, and nothing here is a test result or ranking. Prices and limits change; confirm on the vendor page before budgeting. See our ranking method and the model directory for related profiles.

Frequently asked questions

What is GPT-6 Luna? It is OpenAI's most efficient GPT-6 model, built for focused, high-volume tasks such as extraction, short edits and well-scoped problem solving.

How much does GPT-6 Luna cost? It costs USD 0.10 per 1M input tokens and USD 0.50 per 1M output tokens at list price on 15 October 2026, with cached input at USD 0.01 per 1M tokens.

Can GPT-6 Luna run without reasoning? Yes, Luna supports reasoning.effort set to none, unlike GPT-6 Astra and GPT-6.1 Sol.

What is the GPT-6 Luna context window? It is 1,050,000 tokens, with up to 922,000 input tokens and 128,000 output tokens.

What is the GPT-6 Luna knowledge cutoff? It is 18 May 2026, the latest of the three GPT-6 models.

Does GPT-6 Luna support EU data residency? Yes, OpenAI lists EU data residency with Standard, Fast, Flex and Batch processing.