GPT-6 Luna

gpt-6-luna

GPT-6 Luna is an economy-tier model released by OpenAI on September 22, 2026, on the same day as GPT-6 Sol. It is positioned for clearly defined, high-volume tasks—document summarization, information extraction, and rapid Q&A.

Context window1.0M

ProviderOpenAI

Released2026/09/23

Playground

Pricing

The same model is available through multiple service channels — choose based on latency, reliability and cost.

Prices in $ / 1M tokens

To pick a channel, add a provider field to the request body, for example "provider": { "channel": "direct" }. Valid values are direct / stable / economical; omit it to use the default channel.

Direct

Direct upstream connection — best when you need native behavior and the full context window.

Input context	Input	Output	Cache read	Cache write
≤ 272K	0.10/M	0.50/M	0.01/M	0.13/M
> 272K	0.20/M	0.75/M	0.02/M	0.25/M

Overview

Input

Text Image

Output

Text

GPT-6 Luna API: The High-Throughput Economy Tier of the GPT-6 Series

GPT-6 Luna is an economy-tier model released by OpenAI on September 22, 2026, the same day as GPT-6 Sol. It is positioned for well-defined, high-volume tasks—document summarization, information extraction, and fast Q&A. It is also the GPT-6 model available to ChatGPT Free and Go users.

The core change in this generation is cost: the official input price is half that of GPT-5.6 Luna, and the output price is reduced by 58.3%. Artificial Analysis measured the per-task cost of running the full Intelligence Index dropping from $0.18 to $0.07, a decrease of about 60%. The Intelligence Index score is 37, on par with GPT-5.6 Luna.

Most notable is factuality: OpenAI says that at higher reasoning effort, Luna's factuality can match GPT-5.6 Sol at about one-hundredth of its cost. Third-party data points in the same direction—the AA-Omniscience hallucination rate dropped from 93% to 77%.

SeaWhale AI provides GPT-6 Luna through an OpenAI-compatible API, supporting tool calling, reasoning effort control, streaming output, structured outputs, and image-and-text input.

Get an API key · Model ID: gpt-6-luna

Why Choose GPT-6 Luna

Per-task cost reduced by 60% —— AA Intelligence Index per-task cost is $0.07, compared with $0.18 for GPT-5.6 Luna
Price tier is about 1/20 that of Sol —— in both input and output directions, suitable for covering high-frequency call paths
Hallucination rate dropped significantly —— AA-Omniscience hallucination rate fell from 93% to 77%, and the score improved from -10 to 1
High reasoning effort can match the factuality of the previous-generation flagship —— OpenAI says factuality matches GPT-5.6 Sol at about one percent of the cost
Fast output speed —— AA measured about 157 tokens/second
1.05M token context —— the economy tier also supports a maximum input of 922K tokens and a maximum output of 128K tokens

Core Capabilities

01 High-Volume Summarization and Information Extraction

This is OpenAI's clear positioning for Luna: well-defined, repeatable, high-frequency tasks. Combined with structured outputs and the Batch API, work such as document summarization, field extraction, and classification and labeling can be scaled out at a very low unit cost. AA measured an output speed of about 157 tokens/second, which is relatively fast among reasoning models at the same price point.

Long-document summarization and key point extraction
Structured extraction of fields from forms, contracts, résumés, etc.
Intent recognition and content classification

02 Low-Cost Factual Q&A

Luna shows clear factuality improvements in this generation: the AA-Omniscience score improved from -10 to 1, and the hallucination rate dropped from 93% to 77%. OpenAI's internal factuality evaluation shows that at higher reasoning effort, Luna can match GPT-5.6 Sol's level. For large volumes of simple Q&A, this means there is no need to upgrade to the flagship tier just for reliability.

Customer service and knowledge-base Q&A
Fast fact lookups
Batch generation where fabrication rate needs to be controlled

03 Lightweight Agents and Business Automation

AutomationBench-AA improved from 50.2% to 53.2%, and Terminal-Bench 4.0 improved from 11.6% to 12.6%. In multi-agent architectures, Luna is suited to subtasks such as retrieval, filtering, and simple tool calling, while leaving parts that require deep reasoning to Sol.

Parallel sub-agent execution
Initial screening and ranking of retrieval results
Simple tool-calling loops

04 Coding Assistance

The official DeepSWE v1.1 score at the max setting is 66.6%, only 2.2 percentage points behind GPT-6 Sol's 68.8%. However, the AA Coding Agent Index is 41, 2 points lower than GPT-5.6 Luna. It is suitable for code completion, simple fixes, and batch code processing; for complex engineering tasks, Sol is still recommended.

Batch code formatting and migration
Simple bug fixes
Code comments and documentation generation

Best Use Cases

Use Case	Description
Document summarization	High throughput, low unit cost; the official top recommended use case
Information extraction	Generate structured JSON from unstructured text
Fast Q&A	Hallucination rate down to 77%, suitable for high-frequency simple Q&A
Content classification and moderation	Intent recognition, labeling, compliance checks
Sub-agents	Low-cost execution units in multi-agent architectures
Offline batch processing	Process large-scale corpora with the Batch API

Differences Between GPT-6 Luna, GPT-5.6 Luna, and GPT-6 Sol

Capability	GPT-6 Luna	GPT-5.6 Luna	GPT-6 Sol
Model ID	`gpt-6-luna`	`gpt-5.6-luna`	`gpt-6-sol`
AA Intelligence Index (v4.3)	37	37	48
AA Coding Agent Index	41	43	57
DeepSWE v1.1 (official, `max`)	66.6%	—	68.8%
AA-Omniscience hallucination rate	77%	93%	60%
AA per-task cost	$0.07	$0.18	$1.06
Context window	1.05M tokens	—	1.05M tokens
Max output	128K tokens	—	128K tokens
Knowledge cutoff	May 18, 2026	—	April 20, 2026
Positioning	High-throughput economy tier	Previous-generation economy tier	Cost-effective primary tier

Actual billing is subject to the real-time pricing card at the top of the page.

Frequently Asked Questions

When was GPT-6 Luna released? September 22, 2026, released the same day as GPT-6 Sol. It launched in the API, Codex, and ChatGPT that day, and was also made available to ChatGPT Free and Go users.

What has been upgraded compared with GPT-5.6 Luna? Mainly cost and factuality: per-task cost dropped by about 60%, the hallucination rate fell from 93% to 77%, and business automation (AutomationBench-AA) improved slightly by 3 percentage points. The overall Intelligence Index is unchanged; both are 37.

Are there any areas where it has regressed? Yes. The AA Coding Agent Index fell from 43 to 41, and SWE-Atlas-QnA fell from 49% to 44%; knowledge-work evaluations also regressed—GDPval-AA v2.1 fell from about 1443 Elo to 1367, and AA-Briefcase fell from 1345 to 1299. It is also more verbose: completing the same batch of tasks uses about 23% more output tokens, which partially eats into the price reduction. When a single prompt exceeds 272K tokens, input is billed at 2x and output at 1.5x.

What are the context and output limits? Context window is 1.05M tokens, with a maximum input of 922K and a maximum output of 128K tokens. Input supports text and images; output is plain text. The knowledge cutoff is May 18, 2026, about one month later than Sol.

How should reasoning effort be set? reasoning_effort supports six levels: none, low, medium (default), high, xhigh, and max. For simple tasks such as classification and extraction, none or low is recommended for the lowest latency and cost; raise it when factuality close to flagship level is needed. Note: when using function calling through the Chat Completions API, the official requirement is to set reasoning_effort to none; if you need reasoning and tool calling enabled at the same time, use the Responses API.

How to choose between Luna and Sol? Use Luna for well-defined, repeatable, high-frequency tasks; use Sol for multi-step reasoning, complex coding, or business workflows that span multiple tools. The two have identical APIs and parameters, so you can switch at runtime based on request complexity.

Why Choose SeaWhale AI to Use the GPT-6 Luna API

Cost-optimization friendly —— Shares the same API with GPT-6 Sol and Astra, allowing upgrades or downgrades as needed
OpenAI-compatible API —— Integrate existing code by changing just two lines
Stable under high concurrency —— The platform handles large-volume calls, with no need to build your own proxy
Unified billing —— Centrally manage usage and spending across multiple models

API

API integration

Model IDUse this value as the model in inference requests

gpt-6-luna

API KeyBearer token used to authenticate inference requests

Manage API keys

Base URLOpenAI compatible · /chat/completions

https://api.haijingai.com/v2/

provider OptionalSelects a service channel; omit it and the system picks the default

"provider": { "channel": "direct" }

gpt-6-luna usage examples

SeaWhale AI is compatible with the OpenAI API protocol, so you can call it with the OpenAI SDK or plain HTTP requests. Streaming is enabled by default.

cURLPythonNode.js

curl https://api.haijingai.com/v2/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
    "model": "gpt-6-luna",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hello!"}
    ],
    "provider": { "channel": "direct" },
    "stream": true
  }'
# provider is optional — remove this line to use the default channel

from openai import OpenAI

client = OpenAI(
    base_url="https://api.haijingai.com/v2",
    api_key="<API_KEY>",
)

stream = client.chat.completions.create(
    model="gpt-6-luna",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"},
    ],
    stream=True,
    # Optional: pick a service channel; omit to use the default
    extra_body={"provider": {"channel": "direct"}},
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://api.haijingai.com/v2',
  apiKey: '<API_KEY>',
})

const stream = await client.chat.completions.create({
  model: 'gpt-6-luna',
  messages: [
    { role: 'system', content: 'You are a helpful assistant.' },
    { role: 'user', content: 'Hello!' },
  ],
  stream: true,
  // Optional: pick a service channel; omit to use the default
  // @ts-expect-error provider is a SeaWhale AI extension, not in the OpenAI SDK types
  provider: { channel: 'direct' },
})

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '')
}