Standard Compute
← All models
GPT-6 Luna · OpenAI

What does GPT-6 Luna cost to run in a coding agent?

GPT-6 Luna costs about $0.0041 per average coding-agent request on our measured traffic, at list prices of $0.10 per million input and $0.50 per million output tokens. It suits cheap, well-scoped steps and scores 38 on the Artificial Analysis Intelligence Index. (Source: Artificial Analysis (artificialanalysis.ai))

List prices: OpenAI API pricing, checked 2026-10-13. Request costs: Standard Compute gateway data, 79,301 coding-agent requests from 75 users, Sep 13 to Oct 12, 2026. How we compute this

Run GPT-6 Luna in your agent from $19/month

See plans

What GPT-6 Luna costs per coding-agent request

Cost at list priceGPT-6 Luna
Prompt under 100K tokens50% of requests, 87% of input cached$0.0015
Prompt of 100K to 272K39% of requests, 88% of input cached$0.0039
Prompt over 272K tokens11% of requests, 86% of input cached$0.017
Average request$0.0041
100 requests$0.41
Median session16 requests; a session ends after 10 idle minutes$0.045
A month of regular usemedian of 20 users active on 5+ days$5.83

Each logged request priced at the vendor's list rates for its own token counts, including cache reads and any higher rate for long prompts. Average output: 698 tokens per request.

What to watch for

Luna is weak on long terminal tasks (16.4% on Terminal-Bench 4.0), so a cheap request can turn into more requests and retries. Its input price doubles above 272K tokens. 11% of our requests crossed that line, and they make up 44% of its cost.

Where GPT-6 Luna sits on the Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index bar chart with GPT-6 Luna (max) highlighted, from artificialanalysis.ai, October 13, 2026
Artificial Analysis Intelligence Index v4.3.2, GPT-6 Luna (max) highlighted, downloaded October 13, 2026. Source: Artificial Analysis (artificialanalysis.ai)

About GPT-6 Luna

GPT-6 Luna is the efficiency tier of OpenAI's GPT-6 family, released September 22, 2026 for summarization, extraction and focused coding. It scores 38 on the Artificial Analysis Intelligence Index (max effort) at about 134 output tokens per second, and is the model OpenAI suggests for Codex on the Free and Go plans. It is weak on long terminal tasks (16.4% on Terminal-Bench 4.0), so it works best for well-scoped steps. Its long-context surcharge starts at 272K input tokens, later than Claude Haiku 5.5's 100K line.

Sources: OpenAI model page · Artificial Analysis · Terminal-Bench 4.0

Agents that run GPT-6 Luna well

Cheap, fast subtasks, routing and subagents in OpenAI-based stacks.

GPT-6 Luna FAQ

How much does GPT-6 Luna cost per coding-agent request?
$0.0041 on average at list price, $0.41 per 100 requests, and $0.045 for a median 16-request session. That is 79,301 real coding-agent requests (Sep 13 to Oct 12, 2026) priced at OpenAI's list rates; the median request sent 99,534 input tokens, 87% of input was read from cache, and requests returned 698 output tokens on average.
Does GPT-6 Luna charge more for long prompts?
Yes. Prompts over 272K tokens are billed at $0.20 per million input and $0.75 per million output tokens, on every token in the request including cache reads. 11% of the coding-agent requests we measured were that large.
Is GPT-6 Luna good for AI agents?
Cheap, fast subtasks, routing and subagents in OpenAI-based stacks. On our editorial scoring it rates 8.5/10 for agentic ability and 8.7/10 for output quality. Very low price: $0.10 in and $0.50 out per million.
What is GPT-6 Luna's context window?
GPT-6 Luna has a 1.1M-token context window — large enough to hold sizeable codebases or document sets in a single run. OpenAI is the maker.
Can I self-host GPT-6 Luna?
No — GPT-6 Luna is a proprietary, hosted-only model from OpenAI. You access it through an API (e.g. OpenRouter) rather than running the weights yourself.
What are the downsides of GPT-6 Luna?
The main trade-offs: Weak on long agentic runs (16.4% on Terminal-Bench 4.0); Clear quality step below GPT-6.1 Sol. It's strongest for cheap, fast subtasks, routing and subagents in OpenAI-based stacks.

GPT-6 Luna compared

The agent matters as much as the model. Which agent gets the most out of GPT-6 Luna? →

Every model on this page is included with Standard Compute

Standard Compute

Current frontier models Claude Opus 5.5 · Claude Sonnet 5.5

Your agents never stop.
Your bill never grows.

Frontier models when it counts. Efficient models when it doesn’t. The most intelligence per dollar. One flat bill.

Plans from $19/mo · cancel anytime · 7-day fair refund

7-day fair refund No credit card needed Billing by Stripe1,800+ agent users