Share This Article
GPT-6 comes in three models. Astra is OpenAI’s most capable model and the most expensive. Sol is the everyday option for demanding work at a fraction of Astra’s price. Luna is the fast, cheap model for high-volume jobs. For most marketers, Sol covers daily work, Luna handles bulk tasks, and Astra is worth paying for only on long, high-stakes jobs.
OpenAI didn’t release one GPT-6 this September. It released three, a few weeks apart, and the model picker got more confusing as a result. This guide explains what each model is for, what it costs, where you can actually use it, and a simple way to decide.
GPT-6 models at a glance
| GPT-6 Astra | GPT-6 Sol | GPT-6 Luna | |
|---|---|---|---|
| Released | Early September 2026 | 22 September 2026 | 22 September 2026 |
| Positioned as | OpenAI’s most capable model | Everyday model for demanding work | Fast, low-cost, high-volume model |
| API price (per 1M tokens) | $10 input / $50 output | $2 input / $10 output | $0.10 input / $0.50 output |
| API model ID | gpt-6-astra | gpt-6-sol | gpt-6-luna |
| In ChatGPT | Paid plans (Plus, Pro, Business, Enterprise) | ChatGPT Work and Codex on paid plans | ChatGPT Work and Codex on paid plans; Free and Go users in the desktop app |
| Best for | Long, complex, high-stakes tasks | Reports, analysis, writing, most daily work | Classifying, extracting, summarizing at scale |
Prices are OpenAI’s list prices at the time of writing. OpenAI also discounts cached input tokens heavily, which matters if you send the same long instructions over and over.
How we got here: the September 2026 timeline
- Early September: OpenAI released GPT-6 Astra, first to a limited group and then to paid ChatGPT plans and the API. OpenAI calls it its most capable model and says it leads on computer use, coding and professional work.
- 22 September: OpenAI added GPT-6 Sol and GPT-6 Luna, built with similar methods to Astra but designed to be faster and cheaper. On the same day, Anthropic released Claude Opus 5.5.
The naming follows the pattern OpenAI started earlier in the year, when the first Sol and Luna models appeared in the 5.x generation. The 6-series Sol and Luna cost about half as much in the API as their GPT-5.6 versions.
GPT-6 Astra: the flagship
What it’s for. Astra is the model to reach for when a task is long, has many steps, or costs you real money if it goes wrong. OpenAI positions it as its best model across the board, particularly for computer use (operating software on your behalf), coding, research and professional documents.
Where you can use it. In ChatGPT on Plus, Pro, Business and Enterprise plans, and in the API as gpt-6-astra. OpenAI also lists it on Microsoft Azure and Amazon Bedrock.
What it costs. $10 per million input tokens and $50 per million output tokens in the API, five times the price of Sol.
Good marketing uses.
- Building a full campaign plan from a long brief, research notes and past performance data
- Auditing a large website or ad account export where one missed detail changes the conclusion
- Agent-style jobs where the model works through a sequence of tools or browser steps
GPT-6 Sol: the everyday workhorse
What it’s for. Sol is meant for the demanding work most people do every day: analysis, reports, writing, spreadsheet-to-summary jobs. OpenAI says Sol makes about half as many mistakes as the previous Sol on its internal factuality test, and that on one computer-use benchmark (OSWorld 2.0) it matches Claude Opus 5 at much lower cost.
Where you can use it. In ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and in the API as gpt-6-sol. At launch, OpenAI said Sol and Luna were not yet available in regular ChatGPT conversations, so check your model picker.
What it costs. $2 per million input tokens and $10 per million output tokens.
Good marketing uses.
- Weekly performance reports from exported data
- First drafts of blog posts, emails and landing page copy (which you then edit)
- Turning meeting notes or research into briefs and slide outlines
GPT-6 Luna: the high-volume option
What it’s for. Luna is built for speed and volume, where you run the same simple task thousands of times. Think sorting, tagging, pulling fields out of text and first-pass summaries.
Where you can use it. In ChatGPT Work and Codex on paid plans, in the ChatGPT desktop app for Free and Go users, and in the API as gpt-6-luna.
What it costs. $0.10 per million input tokens and $0.50 per million output tokens, 20 times cheaper than Sol.
Good marketing uses.
- Tagging thousands of search terms, reviews or support tickets by theme
- Pulling product attributes out of messy product descriptions for a shopping feed
- Quick summaries of many short documents before a human looks closer
What the price difference means in practice
Token prices are hard to picture, so here’s a worked example. Say you want to summarize 1,000 customer reviews. Each review plus your instructions is about 500 tokens in, and each summary is about 100 tokens out. That’s 500,000 input tokens and 100,000 output tokens in total.
| Model | Input cost | Output cost | Total (approx.) |
|---|---|---|---|
| Luna | $0.05 | $0.05 | $0.10 |
| Sol | $1.00 | $1.00 | $2.00 |
| Astra | $5.00 | $5.00 | $10.00 |
This is a rough estimate at list prices. Real costs can be higher if the model uses extra “reasoning” tokens before answering (these are usually billed as output), and lower if you use caching or batch processing. The point is the ratio: for simple, repetitive jobs, Luna costs a small fraction of Astra, and you should only pay for the bigger model where it clearly does better.
How to choose: a simple rule of thumb
- Start with Sol for anything a person would normally spend real thought on.
- Move down to Luna when the task is simple, repetitive and easy to check, or when you’re running it at scale.
- Move up to Astra only when Sol gets it wrong on your own tests, or when the task is long, multi-step and expensive to get wrong.
If you’re a ChatGPT user rather than an API user, the choice is partly made for you by your plan. Free and Go users mainly get Luna (in the desktop app). Paid users can pick between them in Work and Codex.
How to test them yourself (in about an hour)
Benchmarks are a starting point, but the only test that matters is your own work. Here’s a quick way to compare the three:
- Pick five real tasks you do every week. For example: an ad copy set, a data summary, a content brief, a customer email reply and a search-term classification.
- Use the exact same inputs and instructions for all three models.
- Record three things for each run: how long it took, what it cost (or how much of your plan it used), and how many things you had to fix.
- Score the output from 1 to 5 on accuracy and on how usable it was without editing.
- Pick the cheapest model that scores a 4 or better on each task.
You’ll often find Luna is fine for classification, Sol handles most writing and analysis, and Astra only earns its price on the longest, messiest jobs.
What about Claude and Gemini?
GPT-6 isn’t the only new option. Anthropic released Claude Opus 5.5 on the same day as Sol and Luna, and Google has said Gemini 4 is in its final training stage. We compare Claude Opus 5.5 and GPT-6 Sol for marketing work in a separate guide, and we’re tracking what’s confirmed about Gemini 4.
Frequently asked questions
Is GPT-6 free?
Partly. Free and Go users can use GPT-6 Luna in the ChatGPT desktop app. Astra and Sol need a paid ChatGPT plan or API access.
What is the difference between GPT-6 Sol and Luna?
Sol is the more capable of the two and suits reasoning-heavy work like analysis and writing. Luna is faster and much cheaper, and suits simple, high-volume tasks like sorting and extracting information.
Is GPT-6 Astra worth the extra cost?
Only for tasks where Sol falls short in your own testing, or where a mistake would be expensive. For most everyday marketing work, Sol is the better value.
Can I use Sol and Luna in normal ChatGPT chats?
Not at launch. OpenAI said Sol and Luna were available in ChatGPT Work and Codex, and not yet in regular Chat. Check your model picker, as this may change.
What are the API model names?
gpt-6-astra, gpt-6-sol and gpt-6-luna.
The bottom line
Don’t default to the biggest model. Use Sol for most work, Luna for volume, and save Astra for the jobs where it proves its worth on your own tests. The price gap between them is large enough that picking the right one for each task matters more than picking the “best” one overall.
For a round-up of everything else OpenAI announced this month, see our DevDay 2026 recap. For ready-made prompts to try on each model, see our prompt templates for performance marketers.
Sources
- OpenAI, GPT-6 Astra: a new generation of intelligence
- OpenAI, Introducing GPT-6 Sol and Luna
- TechCrunch, OpenAI launches GPT-6 Sol and Luna (22 Sep 2026)
- MacRumors, OpenAI’s new GPT-6 Sol and Luna models (22 Sep 2026)
