GPT-6 Astra vs Sol vs Luna in ChatGPT
Astra now ranks above Sol, so older guides name the wrong best model. Here is what each GPT-6 tier is for in ChatGPT and what the same models cost in the API.
The short answer
GPT-6 Astra vs Sol vs Luna is a question about tiers. Astra is the top model in the GPT-6 family. Sol is the middle tier. Luna is the low-cost tier. OpenAI's own latest-model guide for developers uses exactly those positions: it labels Astra "Highest intelligence", GPT-6.1 Sol "Balanced speed, cost, and intelligence", and Luna "Fastest and most cost-effective". OpenAI's individual API model pages word it differently but keep the same order. The Astra page calls it "Our most capable model for the most demanding work." and the Luna page calls it "Our most efficient model for focused, high-volume tasks." Since October 26, 2026, the guide also lists a fourth model, GPT-6.1 Terra, labelled "Fast and efficient". In ChatGPT, Terra is an option in Work and Codex, not in the chat window.
GPT-6 is now in ChatGPT, but not on every plan. OpenAI's deployment safety page for the October release, published October 7, 2026, announced: "We're launching GPT-6 to everyone in ChatGPT, across free and paid plans globally." That is not where things stand today. OpenAI's ChatGPT model help page, checked on November 3, 2026, says: "GPT-6 is rolling out in Chat to eligible Pro, Plus, Enterprise, and Business users." For Free and Go, it says: "GPT-5.6 Luna is becoming the default model for Free and Go users." The safety page also said the new Sol and Luna "will replace GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT", but on Free and Go, GPT-5.6 Luna is still the model.
This guide is for people who use ChatGPT in the chat window. It explains what each tier is for and which plan gets which model. It also gives the price of the same models in the API (application programming interface, the paid developer service). Where OpenAI has not confirmed something on a page we could open, the guide says so.
Why older guides call Sol the best model
The name "Sol" has belonged to more than one model. Before this release, ChatGPT ran GPT-5.6 Sol and GPT-5.6 Luna. OpenAI's safety page for Astra calls it "the most capable model we have ever broadly deployed", and it compares Astra mainly against GPT-5.6 Sol. So a guide written for the GPT-5.6 lineup that calls Sol the best OpenAI model is describing a lineup that is now being replaced on paid plans. In GPT-6, Astra sits above Sol.
The second trap is timing. Some guides from late September 2026 say Sol and Luna are not in ChatGPT yet. GPT-6 has been rolling out in Chat to paid plans since October 7, so those guides are out of date for Plus, Pro, Business and Enterprise. Guides that say every plan already has GPT-6 are wrong in the other direction: OpenAI's model help page lists no GPT-6 model for Free or Go in Chat. Check the date on any guide before you trust its ranking.
The third trap is the version number. GPT-5.6 Sol, GPT-6 Sol and GPT-6.1 Sol are three separate models. When you read a claim about "Sol", look for the version. Without it, the claim could be about any of them.
What is each GPT-6 tier built for?
OpenAI's latest-model guide describes each tier in one line. These are OpenAI's words, quoted directly:
- GPT-6 Astra: "For the most demanding reasoning, coding, and professional work."
- GPT-6.1 Sol: "Near-Astra performance for complex work at a lower cost."
- GPT-6.1 Terra: "For visual reasoning, precise computer use, and clear, natural writing."
- GPT-6 Luna: "Strong performance for focused, high-volume tasks."
In practice, pick the tier by the size of the job. Long reasoning and hard code are what Astra is for. Everyday writing and most coding questions are Sol's job. Short, repeated tasks such as sorting messages or rewriting a sentence fit Luna, because Luna is the fastest and the cheapest tier. Terra is the newest of the four. OpenAI's model help page says "GPT-6.1 Terra is a model option for ChatGPT Work and Codex." and "It is not available in Chat."
OpenAI's model page for Astra lists a context window (the amount of text the model can read in one request) of 1,050,000 tokens, a maximum output of 128K tokens, and a knowledge cutoff of April 30, 2026. A token is a piece of a word, and API prices are counted in tokens.
Which GPT-6 model does my ChatGPT plan use?
This is the question most chat users have. OpenAI's ChatGPT model help page, checked on November 3, 2026, gives the current answer. When GPT-6 is selected in Chat, it says "GPT-6 Sol powers Instant through Extra High where available." In short:
- Free and Go plans use GPT-5.6 Luna. The page says "Free and Go users receive GPT-5.6 Luna" and lists no GPT-6 model for them in Chat.
- Plus, Pro, Business and Enterprise use GPT-6, powered by GPT-6 Sol, in the chat window. Plus gets the Medium and High thinking levels, not Extra High or Pro.
- The "Pro" reasoning option runs on Astra: "GPT-6 Pro, powered by GPT-6 Astra, is available in ChatGPT for Pro $100, Pro $200, Business, and Enterprise plans."
The ChatGPT release notes, also checked on November 3, 2026, tell an older story. The October 7 entry says Plus, Pro, Business and Enterprise use GPT-6 Sol and that Free and Go use GPT-6 Luna, with Free and Go following one day after the paid plans. The model help page does not mention GPT-6 Luna at all, so do not count on it being in Chat on a free plan. For Enterprise, availability depends on admin settings. To check what your own account uses today, open the model menu in ChatGPT.
If you are on a free plan and also wonder about ads, that depends on the plan too, and it is covered in which ChatGPT plans show ads.
Codex and ChatGPT Work run different versions
The new models did not reach every OpenAI product in the same form. In Chat, the GPT-6 option runs GPT-6 Sol. Work and Codex also offer a newer Sol. OpenAI's model help page says: "GPT-6.1 Sol is a model option for ChatGPT Work and Codex." It adds that this option is separate from the GPT-6, GPT-5.6 and GPT-6 Pro options in Chat. OpenAI's Work and Codex help article, checked on November 3, 2026, names four separate models there: "GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Sol and GPT-6 Luna are models for ChatGPT Work and Codex." Astra is there too: "Plus plans include GPT-6 Astra in ChatGPT Work and Codex", and the model help page tells Pro users "Your Pro plan includes GPT-6 Astra in Chat, ChatGPT Work, and Codex." In Codex, "Astra requires Codex CLI version 0.153.0 or newer."
The timing explains the gap. A September 29, 2026 release notes entry brought GPT-6.1 Sol to Work and Codex, starting with Pro users and then expanding to Plus, Business, Enterprise and Edu. The October 7 release changed Chat but, per the release notes, did not change the models behind Work and Codex. The October 7 safety page adds that users of GPT-6 Sol and GPT-6 Luna in Codex and Work "are still using previously released versions". GPT-6.1 Sol sits beside those two as its own option. It does not replace GPT-6 Sol there, and OpenAI's Codex models page also uses gpt-6.1-sol and gpt-6-sol as separate model IDs. The newest addition came on October 26, 2026, when the release notes said "GPT-6.1 Terra is rolling out in ChatGPT Work and Codex." The Codex models page adds: "GPT-6.1 Terra requires Codex CLI 0.159.0 or later."
So "Sol" in Codex can mean GPT-6.1 Sol or GPT-6 Sol, and neither one is guaranteed to be the same build as Sol in the chat window. If a result in Codex differs from a result in chat, check which Sol your Codex session uses; you can pick the model with /model in an interactive session. If you run Codex through your ChatGPT plan, using Codex with a ChatGPT subscription explains how that setup works.
GPT-6 API prices for Astra, Sol and Luna
ChatGPT plans have a monthly price. The API charges per token instead. These are the standard-tier list prices on OpenAI's API pricing page, checked on November 3, 2026. All prices are in US dollars per 1 million tokens, for prompts of 272K input tokens or fewer.
The data behind this chart
[
{
"label": "gpt-6-astra",
"input_usd": "10.00",
"cached_input_usd": "1.00",
"cache_write_usd": "12.50",
"output_usd": "50.00"
},
{
"label": "gpt-6.1-sol",
"input_usd": "2.00",
"cached_input_usd": "0.10",
"cache_write_usd": "2.50",
"output_usd": "10.00"
},
{
"label": "gpt-6-sol",
"input_usd": "2.00",
"cached_input_usd": "0.20",
"cache_write_usd": "2.50",
"output_usd": "10.00"
},
{
"label": "gpt-6.1-terra",
"input_usd": "1.00",
"cached_input_usd": "0.05",
"cache_write_usd": "1.25",
"output_usd": "5.00"
},
{
"label": "gpt-6-luna",
"input_usd": "0.10",
"cached_input_usd": "0.01",
"cache_write_usd": "0.125",
"output_usd": "0.50"
}
]Astra costs $10.00 per million input tokens and $50.00 per million output tokens. Luna costs $0.10 for input and $0.50 for output. Both Sol models sit between them at $2.00 for input and $10.00 for output. The new GPT-6.1 Terra costs $1.00 for input and $5.00 for output, which puts it between Sol and Luna.
Output tokens cost five times as much as input tokens on every model in the table. That matters because a model that writes long answers costs more than one that only reads long prompts.
Cached input is the lower price for input the API has already seen in a recent request, such as a long system prompt that repeats. Cache writes are the price for storing that input the first time. The pricing page also lists Batch, Flex, Fast and Ultrafast tiers with different rates. This guide covers only the standard tier.
Why does the API list two Sol models?
The API has both gpt-6-sol and gpt-6.1-sol. Their input and output prices are the same. The difference in the table is cached input: GPT-6.1 Sol charges $0.10 per million cached tokens, and GPT-6 Sol charges $0.20.
OpenAI's latest-model guide names GPT-6.1 Sol as the balanced tier. It mentions gpt-6-sol only for developers who already use it: "If you already use gpt-6-sol, review the migration guidance before switching to GPT-6.1 Sol." The guide also says gpt-6-sol supports a reasoning effort of none, which GPT-6 Astra and GPT-6.1 Sol do not.
For chat users, OpenAI's model help page says GPT-6 Sol powers the GPT-6 option in Chat, and it lists GPT-6.1 Sol as a separate option for Work and Codex. So do not assume the chat window runs the 6.1 build.
What does one request cost on each model?
Per-million prices are hard to picture. This chart applies the short-context prices above to one example request: 10,000 input tokens and 1,000 output tokens, with no caching. That is roughly a long document plus a one-page answer. The figures are our arithmetic from OpenAI's list prices, not a number OpenAI publishes.
The data behind this chart
[
{
"label": "gpt-6-astra",
"input_cost_usd": "0.10",
"output_cost_usd": "0.05",
"total_usd": "0.15"
},
{
"label": "gpt-6.1-sol",
"input_cost_usd": "0.02",
"output_cost_usd": "0.01",
"total_usd": "0.03"
},
{
"label": "gpt-6-sol",
"input_cost_usd": "0.02",
"output_cost_usd": "0.01",
"total_usd": "0.03"
},
{
"label": "gpt-6.1-terra",
"input_cost_usd": "0.01",
"output_cost_usd": "0.005",
"total_usd": "0.015"
},
{
"label": "gpt-6-luna",
"input_cost_usd": "0.001",
"output_cost_usd": "0.0005",
"total_usd": "0.0015"
}
]The request costs $0.15 on Astra, $0.03 on Sol, $0.015 on Terra and $0.0015 on Luna. One request is cheap on any of them. The gap matters at volume: a script that sends the same request thousands of times a day pays for every one of them, so the tier choice decides most of the bill.
Prompts over 272K tokens cost more
The pricing page defines two bands: "Short context: ≤272K input tokens. Long context: >272K input tokens." Above that line, a higher price applies. For Astra, OpenAI's model page words it this way: "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." The key words are "the full request". The higher rate then covers every token in that request, including the first 272K.
The data behind this chart
[
{
"label": "gpt-6-astra",
"input_usd": "20.00",
"cached_input_usd": "2.00",
"cache_write_usd": "25.00",
"output_usd": "75.00"
},
{
"label": "gpt-6.1-sol",
"input_usd": "4.00",
"cached_input_usd": "0.20",
"cache_write_usd": "5.00",
"output_usd": "15.00"
},
{
"label": "gpt-6-sol",
"input_usd": "4.00",
"cached_input_usd": "0.40",
"cache_write_usd": "5.00",
"output_usd": "15.00"
},
{
"label": "gpt-6.1-terra",
"input_usd": "2.00",
"cached_input_usd": "0.10",
"cache_write_usd": "2.50",
"output_usd": "7.50"
},
{
"label": "gpt-6-luna",
"input_usd": "0.20",
"cached_input_usd": "0.02",
"cache_write_usd": "0.25",
"output_usd": "0.75"
}
]In the long band, Astra input rises to $20.00 and output to $75.00. The table shows the same pattern for Sol, Terra and Luna: input doubles and output rises by half. This is an API pricing rule. The pricing page does not say how ChatGPT plans handle very long conversations.
Does a ChatGPT subscription include API access?
This question comes up often, and the honest answer is short. We did not find an OpenAI page that says a ChatGPT plan includes API usage. We also did not find one that says it excludes it. So this guide does not claim either.
Here is what OpenAI does state. The API pricing page bills per token: "Tokens are billed at the chosen model's input and output rates." It also lists a model called chat-latest under a ChatGPT heading, with its own per-token API price. That is an API model with an API price. It is not a description of a ChatGPT plan. For OpenAI models used through cloud platforms, the page says they "are billed through those services", naming Amazon Bedrock and Microsoft Azure.
If you plan to call the API, budget from the per-token prices above. For the current answer on plans and the API, check OpenAI's help center before you pay for either.
What about message limits and the Pro plan?
This guide does not quote message caps for Astra, Sol or Luna in ChatGPT. OpenAI's model help page, checked on November 3, 2026, describes limits only in words, such as a weekly limit or a daily allowance, with no numbers. It does say "Free and Go users have unlimited everyday text chats, subject to abuse-prevention safeguards", with separate limits for file uploads, images, voice and data analysis. On Pro $200, reaching the weekly GPT-6 Pro limit switches you to GPT-5.6 Thinking at Medium. Secondhand cap figures go out of date quickly.
Pro plan usage has changed twice since late September 2026. A September 29, 2026 entry in the ChatGPT release notes introduces a Pro 500 plan at $500 a month, which has the most included usage of the Pro plans. Another entry the same day says Pro 200 is open to new subscribers again at $200 a month, that its included usage allowances are changing, and that affected users will get an email. An October 29, 2026 entry reverses that change: "We've also canceled the Pro 200 reduction announced in our September 29 email." The same entry says "Pro 100 and Pro 500 plans now include twice their previous Work and Codex allowance", and Pro 200 subscribers on the lower allowance also get twice their previous Work and Codex allowance. The entries do not give the limits as numbers. If usage limits are your reason to choose Pro, read the limits in OpenAI's help center on the day you subscribe. For a plan-level comparison with Anthropic's top consumer plan, see Claude Max vs ChatGPT Pro.
How does GPT-6 compare with Claude on cost?
If you also pay for Claude, the useful comparison is cost. For the per-token side, Claude vs ChatGPT API pricing compares the two APIs. For the monthly side, what a Claude or ChatGPT subscription costs compares the paid plans, and the free plans of Claude, ChatGPT and Gemini covers what you get without paying.
FAQ
Is GPT-6 Sol better than GPT-6 Astra?
No. In the GPT-6 family, Astra is the top tier. OpenAI's latest-model guide labels Astra "Highest intelligence" and GPT-6.1 Sol "Balanced speed, cost, and intelligence". Guides that call Sol the best OpenAI model were usually written for the GPT-5.6 lineup, which GPT-6 has been replacing on paid ChatGPT plans since October 2026. Sol is the better choice when you want lower cost and faster answers for everyday work.
Which GPT-6 model do free ChatGPT users get?
None in the chat window, as of November 3, 2026. OpenAI's ChatGPT model help page says "Free and Go users receive GPT-5.6 Luna" and that GPT-5.6 Luna is becoming their default model, while GPT-6 is rolling out to eligible Plus, Pro, Business and Enterprise users. The October 7 release notes entry said Free and Go would use GPT-6 Luna, but the current model page does not list it. Check the model menu in your ChatGPT app to see what your account uses today.
Why does Codex give different answers from ChatGPT with the same model name?
They may not be the same build. In Chat, the GPT-6 option runs GPT-6 Sol. OpenAI's Work and Codex help article lists GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Sol and GPT-6 Luna as separate models for ChatGPT Work and Codex, and OpenAI's October 7 safety page says the GPT-6 Sol and GPT-6 Luna there are previously released versions. Plus and Pro plans also get GPT-6 Astra there. So Sol in Codex and Sol in the chat window can differ.
How much does GPT-6 cost in the API?
At standard list prices for prompts up to 272K input tokens, Astra costs $10.00 per million input tokens and $50.00 per million output tokens. Both Sol models cost $2.00 and $10.00. GPT-6.1 Terra costs $1.00 and $5.00. Luna costs $0.10 and $0.50. Prompts above 272K input tokens are billed at higher long-context rates for the whole request. OpenAI's API pricing page has the current figures.
Does ChatGPT Plus include OpenAI API credits?
We found no OpenAI page that says a ChatGPT plan includes API usage, and none that says it excludes it. The API pricing page bills per token at each model's input and output rates and does not describe ChatGPT plans. Check OpenAI's help center for the current answer before you rely on either.