Quick answer: GPT-6 Sol and Luna are OpenAI's two cheaper GPT-6 models, released September 22, 2026, sitting under GPT-6 Astra.
Three names, one ladder, one rule: Astra for the hard jobs, Sol for everyday work, Luna for high-volume grunt work.
My honest first read on Sol: "It's nowhere near Fable. Nowhere near Opus 5.5, so a little disappointing." The benchmark table below is why. They roll out in ChatGPT Work and Codex on paid plans. Neither one is in the Chat tab.
And on my own $100 Pro account, the afternoon they launched, they were not in Work yet either. My desktop app had an update downloaded but not installed, so that list came from the older build. I will show you.
Astra is the top rung, Sol is the middle, Luna is the bottom
Here is the whole family on one table, using OpenAI's own words for each model and its API prices.Rung
OpenAI's own line
API price per 1M tokens (in / out)
Leave it on forGPT-6 Astra
"Our most capable model, built for the hardest end-to-end work"
$10 / $50
Hard jobs: long documents, messy research, a real sparring partnerGPT-6 Sol
"Built to power complex coding and agentic workflows"
$2 / $10
Everyday workGPT-6 Luna
"high-volume tasks with a clear goal, like summarizing documents, extracting information, or answering quick questions"
$0.10 / $0.50
Grunt work you repeat a lotPrices and the first two lines are from OpenAI's model docs, read September 22. Luna's line is OpenAI's wording as quoted in TechCrunch's launch report.
Read the price column as a ladder too. Astra costs 5 times what Sol costs per token. Sol costs 20 times what Luna costs.
That's the ladder I'd use: Astra for the hard jobs, Sol for the everyday work, Luna for the grunt work.
If the names still feel like a riddle, you are in good company. On Hacker News the day it launched, dyauspitr asked:"Wtf is GPT-6 Sol, I though GPT-6 is Astra?"Readerium answered with the best two-line explanation I have seen:"Number is generationName is the size (Luna smallest to Astra largest)"That is the system. The thread is worth a scroll if you want to see how many people are confused.
One more thing that trips people up: there is no GPT-6 Terra. The GPT-5.6 family was Sol, Terra and Luna. GPT-6 swapped Terra out and put Astra on top.
Sol and Luna live in Work and Codex, not in the Chat tab
OpenAI's announcement, reposted in its community forum, says it plainly: "GPT-6 Sol and Luna roll out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Both are also available in the API." Launch coverage from 9to5Mac adds that Free and Go users get Luna in the desktop app.
So the Chat tab is not where you will find them. If Work is new to you, what ChatGPT Work is covers it in plain English.
For which plan gets what on which surface, I keep one table on which ChatGPT plans get Astra, and I am not going to duplicate it here.
Now the part the announcement cannot tell you. This is the Work model list on my Pro account on September 22, the afternoon of the launch:My ChatGPT Pro Work tab, September 22, 2026, pre-update: Astra, but no GPT-6 Sol or Luna yet.
And here is the same list twelve days earlier, for comparison:Same account, same tab, September 10, 2026. No Astra, and the names did not even carry the "GPT-" prefix yet.
Two things to take from that. "Rolls out today" and "is in your list today" are different days. And when GPT-6 Sol does land, it is going to sit right next to GPT-5.6 Sol, so check the number, not the name.
If yours is missing too, update the desktop app before you conclude anything about your plan. Mine had an update downloaded but not yet installed when I took that screenshot, so a newer build may already show them. That is the same step OpenAI documented when GPT-6 Astra went missing for people.
Leave Plus on Sol for everyday work and save Astra for the hard jobs
Would I tell a Plus user to leave Work on Sol for the everyday stuff and save Astra for the hard jobs?
Yeah, I would.
Here is the honest part. I got used to using Astra, not gonna lie. But I run it on low, because it burns usage fast.
That is literally a complaint, and it's legit. It's not efficient right now. It burns tokens.
The numbers back me up. OpenAI's own allowance table from September 10 puts Astra at 5 to 45 messages per five hours on Plus, against 10 to 100 for GPT-5.6 Sol. That is roughly double the usage for the same conversation, and I walk through every tier in Astra's usage limits.
That table is for the old Sol. OpenAI has not published ChatGPT limits for GPT-6 Sol or Luna yet.
The same OpenAI help article is also why running Astra on low is not crazy. It says Astra at Low effort can outperform Sol at High effort. That was written about GPT-5.6 Sol, so treat it as a hint, not a rule for the new one.
So my advice is simple: Sol stays selected for the everyday stuff. Astra comes out for the hard job, on low effort first.
Sol's real wins are price and a reliability claim
Sol deserves a fair hearing, and it has two genuine wins.
The first is price. At $2 in and $10 out per million tokens, Sol is one fifth of Astra's API price. An OpenAI spokesperson told VentureBeat these "are permanent prices, not promotional or introductory pricing."
The second is reliability. OpenAI says, per TechCrunch, that "GPT-6 Sol makes about half as many mistakes as its predecessor, reaching Astra-level reliability at much lower cost." That is OpenAI's own internal test, not an independent one.
OpenAI's Tibo (Thibault Sottiaux) went further, in a post the community thread reposts: "GPT-6 Sol and Luna are out. Not only are they a very significant improvement across the board, but also in writing and general 'you know when you try it' quality."
My read, before I have run it: Sol should be good for a lot of the stuff. I expect it to be better than GPT-5.6 Sol, and more efficient. OpenAI's claim points the same way.
OpenAI compared Sol to Opus 5 at medium effort, and Opus 5.5 shipped the same day
Here is where I am a little disappointed.
My opinion, going by the benchmarks and not my own runs: OpenAI basically dropped a cheaper, lower-tier model. It's barely at Opus 5 level.
That read holds up on OpenAI's own chart, with one caveat you should know about:Model
Effort setting
OSWorld 2.0 (computer use)
Cost per taskGPT-6 Sol
xhigh
60.5%
About 80% lowerClaude Opus 5
medium
60.3%
BaselineThese are OpenAI's numbers, as reported by VentureBeat on September 22.
Look at the effort column. Sol's effort dial runs none, low, medium, high, xhigh, max, so xhigh is its second-highest setting. Opus 5 was at medium.
Sol had to run near its top setting to tie Opus 5 at medium, and by OpenAI's math it still did that for about a fifth of the cost.
Then Anthropic released Claude Opus 5.5 hours before OpenAI's announcement. Anthropic says Opus 5.5 performs at the level of Fable 5.1 on most work, at $4 in and $20 out per million tokens. OpenAI's material has no Opus 5.5 comparison.
In fairness to Sol, it is still half of Opus 5.5's price per token. I compare Opus 5.5 with Fable properly in Opus 5.5 vs Fable.
The way I see it, 5.5 is a huge leap. You literally got a better model and it's cheaper.
Sol is a sidekick for grunt work, not my main model
Does Sol feel like a real option to me? To be honest, no.
I think it's gonna be a good sidekick for research and the grunt work I just want to burn usage on. If I want to tinker around and brainstorm, I'm still gonna use Astra.
So I have to find a use case for it. That is the honest status on day one: I have not run GPT-6 Sol or Luna on a real job yet, and it was not even in my list when I wrote this, on an app build one update behind.
I also have a complaint about the model count, and then I have to take half of it back.
I think this whole game of having five different models makes it seem like you have more options, when really you've got one or two good things. That's what I like Anthropic for right now, because it's not complicated. You've got Haiku, Sonnet, Opus and Fable.
So they do have a lot of models too, actually. Never mind.
But I rarely use Haiku. I rarely use Sonnet. I'm just using Opus and Fable. If that is where you are too, how I run Claude is the setup I actually use.
What nobody knows yet
A few things were not public when I wrote this, and I would rather say so than guess:ChatGPT limits for GPT-6 Sol and Luna. OpenAI has published no message numbers for either one.
When the Chat tab gets them. The announcement only names Work and Codex.
How much Luna a Free account gets. Launch coverage says Free and Go users get Luna in the desktop app, with no number.
Whether GPT-6 Sol still loses to Astra on Low. OpenAI's effort advice was written for GPT-5.6 Sol.I log plan and limit changes with dates on the AI plan tracker, and the box at the bottom of this page emails you when one moves.
Which one to pick depends on what you pay forIf you are on Plus: leave Work on Sol for everyday tasks once it shows up. Pull out Astra for the hard job, on low effort first, because it cost roughly double the usage of the old Sol.
If you are on Pro: you have room for Astra, and I still run it on low. Send Sol the bulk research you would rather not spend Astra on.
If you are on Free or Go: Luna in the desktop app is your only GPT-6 door. Use it for summaries, pulling details out of a document, and quick questions, which is exactly what OpenAI built it for.
If you also pay for Claude: Opus 5.5 is the one to beat now, and OpenAI did not even compare Sol to it. Treat Sol as a cheap sidekick, not a switch.Frequently asked questions
What is GPT-6 Sol?
GPT-6 Sol is OpenAI's middle GPT-6 model, released September 22, 2026. It sits under GPT-6 Astra and above GPT-6 Luna.
OpenAI's developer docs describe it as built to power complex coding and agentic workflows, and it costs $2 per million input tokens and $10 per million output tokens in the API, one fifth of Astra's price.
What is the difference between GPT-6 Sol and Luna?
Sol is the everyday model for real work. Luna is the small, cheap one for high-volume tasks with a clear goal, which OpenAI lists as summarizing documents, extracting information, or answering quick questions.
In the API, Sol costs $2 in and $10 out per million tokens, and Luna costs $0.10 in and $0.50 out, so Sol is 20 times the price of Luna.
Is GPT-6 Sol better than GPT-6 Astra?
No. Astra is the top of the GPT-6 ladder and Sol is the middle rung.
OpenAI says Sol reaches Astra-level reliability on its internal factuality test at much lower cost, but Astra is still the model OpenAI builds for the hardest end-to-end work. Use Sol for everyday tasks and save Astra for the hard jobs.
Where can I use GPT-6 Sol and Luna in ChatGPT?
OpenAI says both roll out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, and not in the Chat tab. Launch coverage (9to5Mac, VentureBeat) adds that Free and Go users get GPT-6 Luna in the ChatGPT desktop app.
On launch day afternoon, my own Pro account's Work list still showed GPT-5.6 Sol and Luna, on a desktop app with an update downloaded but not yet installed, so update the app first if yours is missing them.
Is GPT-6 Luna free?
Partly. Launch coverage from 9to5Mac and VentureBeat says Free and Go users get GPT-6 Luna in the ChatGPT desktop app, and OpenAI's help center had not caught up on launch day. Sol is for paid plans only. OpenAI has not published how many Luna messages a free account gets.
What happened to GPT-6 Terra?
There is no GPT-6 Terra. The GPT-5.6 family had three sizes, Sol, Terra, and Luna. The GPT-6 family announced so far has three different names: Astra on top, Sol in the middle, and Luna at the bottom.
Where this comes from
The facts come from OpenAI's docs and announcement, plus TechCrunch and VentureBeat, read September 22, 2026. The screenshots are my own Pro account, and the opinions are mine.
I have not run GPT-6 Sol or Luna yet. My list still showed the 5.6 versions, with an app update downloaded but not installed.
Published September 22, 2026. Last reviewed September 22, 2026.
Quick answer: ChatGPT Work is the mode that hands you a finished file instead of a reply. It is for longer, multi-step jobs that end in a document, spreadsheet, deck, report or Site.
It is still rolling out, and it draws on the same usage pool as Codex. Open it when the output is a deliverable. Stay in chat when the output is an answer.
Chat asks, Work delivers. That's my read on Work, and the part nobody leads with is what the delivery costs you.
Chat asks, Work delivers, and Codex is a third thing
OpenAI splits the three by job, and their own wording is the cleanest definition on the internet right now.Experience
What OpenAI says it is for
What you get backChat
Ask questions, search, brainstorm, or get quick conversational help
An answerWork
Research a topic, analyze information, or create a document, spreadsheet, presentation, report, or Site
A finished fileCodex
Write or debug code, run tests and commands, review changes, or work with a repository
Changed codeVoice
Talk naturally with Work or Codex in the desktop app and ask it to start or coordinate tasks
A running taskThe plainest way I can put it: Work is a chat with hands. It can actually edit your files, open things, delete things. If you have used Claude's file-handling mode, it is the same shape of tool.
lifeandwork.blog, one of the small sites ranking on this exact framing, says it the same way: "In chat mode, it behaves like a conversational partner for ideas, questions, and quick explanations. In work mode, it becomes a structured assistant for drafting, planning, reviewing, and managing tasks."
OpenAI's own two pages disagree about whether Work is free
I went looking for a clean answer to "which plan has it" and did not find one. I found two OpenAI properties saying different things on the same day.
OpenAI's developer docs, read September 18, 2026, open with this: "ChatGPT Work and Codex are included in your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan."OpenAI's help centre, read the same day, says Work is available on ChatGPT web and mobile for eligible paid plans, and in the desktop app when it is included for your plan and workspace.
Free is in one list and absent from the other. I'm not going to pick a winner and call it settled. I can't tell which page is stale, and every other article on this confidently picks one of the two.
What both agree on is the part that actually explains your experience: Work is still rolling out. The help centre says it is gradually rolling out to eligible accounts, and that if you do not see it yet, your account may not have access yet.
So if Work is missing from your ChatGPT, you have not done anything wrong and there is no setting to find. Check your own account, not a pricing table. That goes for this page too. Trust what's in front of you.
One practical note before you go upgrading: new sign-ups and upgrades to the ChatGPT Pro $200 plan are currently paused. Our AI plan tracker follows that sort of change.
Where Astra sits, and why your usage vanishes
Every capability tour skips this section. It's the most useful one on the page.One sentence from OpenAI's own page does most of the work: "Plus plans include GPT-6 Astra in ChatGPT Work and Codex."
If you're on Plus, that changes how you should look at Work. It's more than a different way to ask.
On the $20 plan, Work and Codex are the two surfaces where the heavy model reaches you, and ordinary Chat is not one of them. Which plan gets Astra where is Astra by ChatGPT plan, so I won't redo it here.
The line that matters for this page comes next: "Astra can use your allowance faster than GPT-5.6 Sol. How much usage you get depends on the task, input and output size, reasoning settings and Fast mode."
That is OpenAI saying, in their own words, that the model you pick moves the meter. They do not publish a multiplier, and neither will I. Any "Astra burns 5x faster" number you find online is someone's estimate. The detail lives on Astra usage limits.
What I can add is what it's like to use. Astra burns fast enough that I changed my habits. I use it every day, I have run out of usage more than once, and I will not put a firm percentage on it because it swings with the task.That is my own usage and billing pane from September 5, 2026. The meter is a real number, and it goes down while you watch.
Here is what I actually hand to Work
The rule I use is embarrassingly simple: if it is a question, it is chat. If it is a job, it is Work.
Quick question, one answer, done? Chat. "There's a deal on this TV, which one should I get?" That is a chat question and it would be a waste of an agent.
Building something, reviewing my own system, going through my analytics, pulling my YouTube live data apart? That is Work. In my case a lot of that runs through Codex specifically, because my work ends in files and scripts, but the shape is the same.
The test is the output. If what you want back is a file, open Work. If what you want back is a sentence, stay in chat.That is the Work composer on my own account on September 10, 2026, with the model chip showing what is loaded. It is the thing you will actually be looking at, and no competitor page shows it to you.
The first job I would hand it, on camera:The 60-second version, from my Shorts: What to hand ChatGPT Work first (OpenAI's own advice).The expensive lesson: I tried to build a Pokemon game
Here's the honest part, because a capability tour won't give you one.
I pointed Astra at building a Pokemon-style game. It got somewhere. It did not get close to finished. And it ate a serious chunk of my usage doing it.That screen is real and it is as far as it got.
Could I have done it better? Yes, and you can learn from my mistake. I was inefficient. I got too creative, I went all over the place, and I never gave it tight direction.
If I had said "clone the basic thing and keep it simple" I would have spent a fraction of the usage and got further.
This doesn't mean "Work is bad at games". It means Work spends real money-equivalent on vague instructions. Chat doesn't, because chat is not out there doing forty steps on your behalf while you watch.
Give Work a scoped job with a named output. Save the exploring for chat.
TechRadar ran the opposite experiment, handing Work a pile of life-admin chores, and their headline verdict was that it handled them like a pro. That matches what I've seen. Give it a boring job with clear limits and it does well.
When to stay in chat
Every page about Work lists what it can do. This is the other list.
Stay in chat when:You want an answer, not a file.
You are still figuring out what you want. Doing that thinking in Work gets expensive.
The task is one step. An agent that plans, checks and revises is extra work you pay for.
You are on Plus and the job does not need Astra-grade reasoning. OpenAI's own guidance is that Astra moves the meter faster.
You just want speed. OpenAI's framing for Chat is "fast, conversational assistance and everyday questions", and that is still the right tool for most of the day.And no, regular ChatGPT is not going away. OpenAI's FAQ answer is one line: "The fast, familiar Chat experience is still available." On desktop you switch with the toggle at the top of the page. On mobile it is the dropdown at the top of the screen.That is the chat-side picker on my account, and it is a different set of choices from the Work composer further up this page. It's the same subscription with two different menus.
Work vs Codex vs Claude's version, in one tableWhat it is for
Where it runs
Go deeperChatGPT Work
Documents, spreadsheets, decks, reports, Sites
Web, mobile, desktop; local files in the desktop app only
This pageCodex
Code, tests, commands, repositories
CLI, IDE, web, desktop, iOS
Codex vs Claude CodeClaude's file mode
The same shape of job on Anthropic's side
Desktop and web
Cowork vs ChatGPT WorkOne note on that last row: Anthropic has folded Cowork back into Claude itself, with their own docs now carrying the line "Claude Cowork is now just Claude." The feature is still there. Only the product name went away.
So should you open WorkIf you are on Plus and have never opened it: do, once, on a job with a clear deliverable. Work and Codex are also where Astra reaches you on that plan.
If you cannot find it: you are not missing a setting. It is still rolling out.
If you are on Free: check before you assume. OpenAI's docs list Work as included on Free and OpenAI's help centre does not, so your own account is the only reliable answer. Either way you have Codex, which is on every ChatGPT plan, covered on Codex vs Claude Code.
If your usage keeps vanishing: it is probably the model, not the mode. Read Astra usage limits and get deliberate about when you reach for the big one.
If you mostly ask questions: stay in chat and stop feeling behind. Work is for a different kind of job.The Claude Guides hub has the equivalent map on the Anthropic side if you run both like I do.
Common questions
What is ChatGPT Work mode?
Work is an agent mode inside ChatGPT for longer, multi-step tasks that end in a finished deliverable. OpenAI describes it as researching a topic, analyzing information, or creating a document, spreadsheet, presentation, report or Site. Chat, by contrast, is for quick conversational help.
Is ChatGPT Work free?
OpenAI's own two pages disagree, as of September 18, 2026. The developer docs list Work as included on Free, Go, Plus and up. The help centre says eligible paid plans. It is also still rolling out, so check your own account rather than either page.
Why can I not find Work in my ChatGPT?
Because it is still being rolled out. OpenAI's own wording is that Work is gradually rolling out to eligible accounts and that availability will continue to expand. If it is missing, your account may simply not have access yet.
Does Work use more usage than chat?
Work follows the same usage structure as Codex, so it draws on the same agentic allowance. The larger variable is the model: OpenAI says Astra can use your allowance faster than GPT-5.6 Sol, without publishing a rate.
Can Work open files on my computer?
Yes, in the desktop app, where you open a local folder or project and grant access only to the files a task needs. Work on web and mobile cannot directly access files on your computer.
Where this comes from: OpenAI's own ChatGPT Work and Codex help article, read and screenshotted September 18, 2026, two days after OpenAI updated it. I pay for ChatGPT and use Work and Codex daily. The Pokemon game and the usage screens are mine, taken September 10. No usage multiplier, because OpenAI publishes none. Last reviewed September 2026.
Quick answer: these two do not compete; they sit on different layers. GPT-6 Astra is a model you hand hard work to.
Grok Bot is a set of agents with their own computer that run standing jobs while you are asleep. I pay for both and they have never once contended for the same task.
If you came for the model question: Grok 4.7 shipped September 21, and the independent Artificial Analysis index still has Astra ahead, 53 to 46.
Does 4.7 close the gap? No. It does not compete with it.
You can put the benchmarks up and laugh. This is not even close.
Every other result ranking for this pairing is framed around benchmarks or API prices: tokens per second, dollars per million, intelligence index. Useful if you are building software. Useless if you have two subscriptions and one Tuesday.
So here is the version for someone with two subscriptions and one Tuesday. The model table first, because that is what changed this week.
Grok 4.7 vs Astra, the model question in one tableGPT-6 Astra
Grok 4.7
SourceIntelligence Index v4.3.2
53 (max effort)
46 (xhigh effort)
Artificial Analysis, read Sep 22GDPval
1542
1695
xAI's own launch chartTerminal-Bench 4.0, xAI's number
Not shown
37.6%
xAI's launch tableTerminal-Bench 4.0, The Decoder's number
60%
26%
The Decoder, Sep 21API price per 1M tokens (in / out)
$10 / $50
$2 / $6
Artificial Analysis; Grok's matches xAI's launch pageCost to run one index task
$0.82 (low) to $3.26 (max)
$2.73 (high) to $3.74 (xhigh)
Artificial Analysis; effort settings differContext window
1M tokens
500K tokens
Artificial AnalysisWhere you can use it
ChatGPT Work and Codex on paid plans
Cursor, Grok Build, Grok API, third-party tools
OpenAI plan pages; xAI's launch pageRead the GDPval row carefully. It is the one chart where Grok 4.7 beats Astra, and it is the chart xAI chose to print. xAI's own Terminal-Bench table leaves Astra out entirely.
Grok is cheaper per token: five times on input, about eight times on output. It is not clearly cheaper per job. On Artificial Analysis's cost to run a task, Grok 4.7 at high effort lands at $2.73 and Astra at low effort at $0.82, because Grok burns far more tokens to finish.
Those are different effort settings, so read it as a caution, not a verdict.
Even Elon Musk does not call it the leader. His launch-day post, via the Musk post archive: "Grok 4.7 places @SpaceXAI as third, after Anthropic & OpenAI, for agentic coding."
And here is the bigger point for me. Astra doesn't even compete with Fable 5.1, let alone the new Opus 5.5. That should tell you where we're at, right?
That last part is my take, not a scoreboard. On Anthropic's own Opus 5.5 table, Opus 5.5 actually trails Astra on two rows: AutomationBench (40.0 vs 41.4) and Terminal-Bench-Science (58.7 vs 64.6). And The Decoder reports Fable 5.1 and GPT-6 tied at 53 on the same Artificial Analysis index.
None of it should move you much, for one reason: you are on a plan, not an invoice. Price per million tokens describes a bill you will never receive.
Where you can actually use Grok 4.7 today
xAI's launch page says it plainly: "Grok 4.7 is available today in Cursor and Grok Build. It is also available through the Grok API, third-party coding harnesses, and model routers and cloud platforms."
xAI's launch page does not list the Grok app in that availability line. So if you pay for SuperGrok and open the chat app, do not assume the new model is the one answering.
Grok Build is the door every Grok subscriber has, because it has been on every plan, free included, since August 19. I've been running it in Grok Build, the one place I've confirmed I'm on 4.7.
The first thing I handed it there, a script to help me find good business ideas, failed. That is all I have really tested it on.
Grok Bot is a separate question. xAI's launch page does not say which model Grok Bot runs, and Grok Bot itself can't confirm it, so I am not assuming the upgrade reached it.
For the full launch breakdown, here is what Grok 4.7 actually is. For what your plan actually buys you, see how Grok's usage pool works.
And if your real question is Grok Bot vs Astra, you are not choosing between two models anyway. You are choosing between a model and an employee. A better model does not change that split.
The category error that decides everythingGPT-6 Astra
Grok BotWhat it is
A model
Named persistent agentsWhere it runs
Your ChatGPT session, in Work or Codex
Its own cloud virtual machineWhose logins
Yours, in your browser
Its own, authenticated onceWhose files
Whatever you give the session
Its machine, under /workspaceWhen it works
While you are there
On a schedule, without youWhat happens at 3am
Nothing
The job runsHow it is billed
Inside your plan allowance
Weekly allowance, then reported overageRead the last two rows again. That is the entire decision.
Astra is the better thinker in this pairing. But it has never once told me something while I was doing something else.
What Astra actually ran for me
Astra shipped September 3, 2026. OpenAI describes it as state-of-the-art on computer use, browsing, software engineering, cybersecurity, science and professional work, with a context window over a million tokens.
In my week it does two things.
Research, and challenging Claude. It is genuinely good at pushing back on Fable when I want a second opinion on something I already believe. That is a real job and it is worth paying for. Smart enough that the disagreement is worth reading rather than annoying.
And this week, a game. The demos of people building games with Astra are real and I wanted in, so I spent a weekend on a Pokemon-style game. The graphics it generated stunned me. The game did not get built.
Between that and my first week of running maximum effort on everything, I burned through two full allowances and produced something ugly.
That is not a knock on the model. I did not ask it the right way and the learning curve is real. But I want it next to the demo reels, because the demo reels are all successes.Astra in ChatGPT Work on my $100 Pro account. On Plus, Astra shows up only in Work and Codex.
The honest cost note: at $100 I feel the ceiling for the first time. With GPT-5.6 Sol (the older Sol, not the GPT-6 Sol that shipped September 22) I felt it a little, but it was manageable. This is not.
If you want to build something serious with Astra, my read is you want the $200 plan. At $100 you miss out.
One catch: OpenAI paused new sign-ups and upgrades to the $200 plan on September 10, and its help center still said so on September 22. Existing $200 subscribers keep renewing.
What Grok Bot actually ran for me
Grok Bot launched August 11, 2026. Named agents, each with standing instructions, sharing one cloud machine per account along with its logins and files.
I use it for orchestration. Analytics, publishing, intelligence, all happening in the background.
Every day, my inbox. Starting at 9am, about three times a day, a bot checks and manages it. I do not open it to triage. I open it to read what is left.
Every day, my numbers. A folder of bots scrapes my own analytics roughly three times a day and writes to a ledger. Shorts, newsletter, LinkedIn, comments. Wins and losses get logged rather than remembered.
I am no longer posting blind, which sounds small and changed how I work more than any model upgrade this year.
Anything that drops in my space. A news scout watches X. Yesterday it told me OpenAI had shipped a new image generation model, and it was right that it mattered.
We are now looking at integrating it into our YouTube thumbnails. That is a bot changing what I did with my afternoon.
And the sniper job. When I see a tweet I want a take on, I dump it into Grok Bot. It reads it instantly because it lives where the tweets are.
Pasting that into a Claude chat is a headache. I do not want to open a terminal just for that.
This is the fastest path from "interesting" to "what do you think."Inbox Hawk: three passes a day, pings me only when a human is needed. Astra has no slot for that.
The billing asymmetry nobody prints
This is the most practical difference on this page and I have not seen it anywhere else.
ChatGPT stops you. When your allowance runs out, you wait, use a banked reset, or buy credits deliberately. The worst case is frustration.My ChatGPT meter, September 10, 2026: a ceiling, a reset date, $0 balance, auto-reload off. Work just stops.
Grok Bot bills you. Each eligible plan includes a weekly Grok Bot allowance, and extra usage is reported to be billed from model and token cost with no bot-specific spend cap yet. A scheduled agent running unattended with no ceiling is a different category of risk from a subscription that simply stops.
One of these fails by stopping. The other can fail by spending. Watch the second one for a month before you trust it with a schedule.
On price itself, the record needs correcting. Grok Bot is included from SuperGrok at $30, per xAI's pricing page as of September 10, 2026.
Secondary coverage still ties it to SuperGrok Heavy bundling at around $300, and Reworked reported exactly that at launch. Whatever was true then, the page today says otherwise. It is also, notably, not listed as included on Business or Enterprise.xAI's pricing page, September 10, 2026: Grok Bot is on the three individual tiers, not the business ones.
If you can only pay for one
I asked myself the real version of this: if I had to cancel one tomorrow, which goes?
Astra goes. And I want to be clear that this is not a verdict on quality, because Astra is great and I would miss it and I would have to work out what to do about that.
Which sounds like it contradicts my own buying order above, and it does not. If you have neither, buy the model first, because a model is useful on day one and an agent is useful once you have taught it a job. I have already done that teaching. That is the whole difference.
Grok Bot is serving a more important purpose right now. Tracking the analytics, keeping things updated, doing the things I would otherwise not do. Losing Astra costs me a very good research partner and a sparring opponent. I would feel that.
Losing Grok Bot means a category of work simply stops happening in my business.
The way I keep describing the difference: a conventional automation is like a teddy bear you have to move yourself. The moment you stop moving it, it dies. Occasionally it gets up on its own to do a boring chore, and sometimes it trips and falls and stays down.
Grok Bot plays on its own, walks more smoothly, and when it falls it can get up or call one of the other bots for help.
That is a feel, not a benchmark. It is also why I would cancel the smarter one first.
So which do you buy?You mostly ask questions and want better answers. Neither of these. You want a good chat plan, and that is ChatGPT Plus or Claude Pro.
You have hard, long, multi-step work. Astra, on ChatGPT. And if you intend to build something serious with it, budget the $200 tier, because $100 has a ceiling you will meet. New $200 sign-ups have been paused since September 10, so check before you plan around it.
You have recurring jobs you do at the same time every day and resent. Grok Bot, from $30. Start with one bot, not nine.
AI does real work in your week. Both, in that order. And honestly, for a reason beyond capability: running two vendors means one closed meter does not end your day.There is no AI that does it all. If you are running a business on this, do not rely on one. If I were making real money from it, I would be on the top tier of all of them and consider it a no-brainer, because it makes the money back.
The neighboring decisions: what GPT-6 Astra actually is, which ChatGPT plans get Astra, what is Grok Bot for the week-one field test, and Grok Bot use cases for the full roster. If your comparison is really against Claude's assistant, that is Grok Bot vs Claude Cowork. More at Claude Guides.
Questions people ask
Is Grok 4.7 better than GPT-6 Astra?
Not on the independent numbers. Artificial Analysis's Intelligence Index v4.3.2 has GPT-6 Astra at 53 (max effort) and Grok 4.7 at 46 (xhigh), read September 22, 2026. Grok 4.7 wins on price per token ($2/$6 per million vs $10/$50) and on xAI's own GDPval chart, but it is not clearly cheaper per job.
My take: it is not even close.
Is Grok Bot better than ChatGPT?
They are not the same kind of thing, so the question does not resolve. GPT-6 Astra is a model you send work to inside a session.
Grok Bot is a set of named agents with their own cloud machine and their own logins that run standing jobs whether or not you are there. Comparing them on benchmarks compares the wrong layer.
Can GPT-6 Astra do what Grok Bot does?
Partly, inside a session you started. What it does not do is wake up at 9am on its own, log into your accounts, produce something, and write it somewhere for you to read later. That unattended, scheduled, persistent shape is the thing you are actually buying with Grok Bot.
Do I need both?
If AI does real work in your week, probably yes, and in that order: the model first, the agent second. They never contend for the same job in my week. The model handles thinking and hard multi-step work. The agent handles standing jobs I would otherwise do at the same time every day.
Which one should I cancel?
For me, today, Astra goes first. Not because it is worse, but because Grok Bot is now doing something structural: tracking my analytics and keeping things updated without me. Losing Astra costs me a very good research and sparring partner. Losing Grok Bot means a set of jobs simply stops happening.
How much does each cost?
Astra rides on your ChatGPT subscription, so Plus at $20 or Pro at $100 or $200, with usage inside your plan allowance. OpenAI paused new sign-ups and upgrades to the $200 plan on September 10, 2026, and its help center still said so on September 22.
Grok Bot is included from SuperGrok at $30 per xAI's pricing page. The difference that matters is that a ChatGPT plan simply stops you at the ceiling, while Grok Bot overage is reported to bill from model and token cost with no bot-specific spend cap yet.
Is Grok Bot just Grok chat?
No. Grok chat is a chat window. Grok Bot is a separate product launched August 11, 2026, where named persistent agents run on a cloud virtual machine, one per account, shared across your bots, with their own logins and files.
Where this comes from
The Grok 4.7 numbers come from xAI, Artificial Analysis and The Decoder, read September 22. Grok Bot pricing is xAI's page from September 10, and the screenshots are my own paid accounts.
My only 4.7 run was one failed job in Grok Build, so the verdict up top is my read, not a test.
Correction logSeptember 10, 2026. Published. Grok 4.7 was unreleased, so I kept it off the page on purpose.
September 22, 2026. Grok 4.7 shipped September 21, so I added the model table and where you can use it. Replaced the old line that had Astra and Grok 4.6 tied at 61 with faster Grok: that was Artificial Analysis's older index, and today's v4.3.2 has Astra 53, Grok 4.7 46, and Astra faster (61 vs 47 tokens a second). Added the $200 Pro sign-up pause, labeled the GPT-5.6 Sol line as the older model, and put the FAQ on the page. Re-checked xAI's launch page, Artificial Analysis, The Decoder and OpenAI's help center that day.Published September 10, 2026. Updated September 22, 2026 after Grok 4.7 shipped, with the details in the correction log above.
Quick answer: there are two clocks and two separate pools, and OpenAI does publish numbers.
Per its help center, GPT-6 Astra's estimated messages per five-hour window are 5 to 45 on Plus, 25 to 225 on Pro 5x, and 100 to 900 on Pro 20x. A weekly limit sits on top of that, and you need allowance in both to keep going.
So I went and screenshotted the actual table, dated. And from my own $100 plan: I've burned two resets already. The first time I was just a dummy, I was using Ultra for everything.
The finding that matters most is not any single number. It is the ratio sitting inside the table: Astra costs you at least double what GPT-5.6 Sol costs you, at every tier.
The numbers OpenAI actually publishesOpenAI's help article on Astra usage in Work and Codex, captured September 10, 2026, two days after its last update.
The caveat OpenAI prints directly above that table deserves to be read out loud, because every page quoting these numbers drops it:The table below shows estimated local messages per five-hour period. These are not fixed message limits. Actual usage varies by task, model and settings, and weekly limits may also apply.So the honest version of "how many Astra messages do I get" is: somewhere between 5 and 45 on Plus, and which end of that range you land on is decided by what you ask it to do.
Now read the same table for the ratio inside it:Model
Plus
Pro 5x
Pro 20xGPT-6 Astra
5 to 45
25 to 225
100 to 900GPT-5.6 Sol
10 to 100
50 to 500
200 to 2,000GPT-5.6 Terra
25 to 200
125 to 1,000
500 to 4,000GPT-5.6 Luna
250 to 2,000
1,250 to 10,000
5,000 to 40,000Astra is about half of Sol on every row. The bottom of each range is exactly half: 5 against 10, 25 against 50, 100 against 200. The top of each range is a little under half: 45 against 100, 225 against 500, 900 against 2,000.
That is the whole cost story in one line: the same conversation costs you at least twice as much on Astra, and up to about 2.2 times as much.
And if you drop to Luna for the routine stuff, you get roughly 45 to 50 times the messages you would get on Astra.
One thing the column headers will not tell you: Pro 5x is the $100 plan and Pro 20x is the $200 plan. OpenAI labels them by multiplier, your credit card labels them by price, and nobody maps the two for you.
Two clocks, and you need both
This is the mechanic that makes limits feel random when they are not.Same article, same capture date. The shared-pool sentence and the two-window rule, in OpenAI's own words.Depending on your plan, usage limits may apply over a five-hour window and a weekly window. Where both apply, you need allowance remaining in both to continue.The five-hour limit determines how much usage is included in each window. A new window starts when you send your first message in Work or Codex after the previous one ends. It is not a fixed clock on the wall; it starts when you do.
The weekly limit controls how much included work you get over the weekly usage period.The consequence, in OpenAI's words: "You may reach the five-hour limit before five hours have passed, even if you still have weekly usage remaining."
That is why the wall feels arbitrary. You still see plenty of weekly allowance in Settings, so you assume you are fine, and then a single heavy task closes the five-hour door.
Two pools, and Chat is not one of them
The second confusion is which meter you are draining.Surface
Pool
Who gets Astra thereChatGPT Work
Shared Work and Codex allowance
Plus, Pro, Business Standard and PremiumCodex
Same shared Work and Codex allowance
SameChatGPT Chat (as GPT-6 Pro)
Separate Chat message limits
Pro, Business, Enterprise. Not PlusOpenAI's wording: "Work and Codex share the usage allowance included with your plan." And separately: "In Chat, GPT-6 Pro is powered by Astra and is available on eligible Pro, Business and Enterprise plans. Its message limits are separate from the shared Work and Codex allowance and vary by plan."
So running Astra hard in Codex does not eat your Chat allowance. It does eat your Work allowance, because those two are the same bucket.
One more rule that catches people mid-task: "Switching models does not restore allowance in a shared usage pool." Dropping from Astra to Luna after you have hit the wall does not open the door. It only slows the next burn.
The setting that matters more than the model
Here is the paragraph I think is the most valuable thing OpenAI has published about Astra, and it is filed under a settings heading where nobody will read it.Same article, same capture date. Read the bolded sentence twice.Lower effort does not mean lower capability across models. For example, Astra at Low effort can outperform Sol at High effort. If you've been getting good results with Sol at High, try Astra at Low or Medium as a starting point.And the warning attached to the top setting:Higher effort can use more of your allowance and does not always produce a better result.If you take one action from this page, it is that one. If Astra feels like it is draining you, check whether you are running maximum effort on tasks that never needed it. That was my problem.
My own advice lands in roughly the same place. The way I put it: use the top setting to plan the hard thing, then drop down to execute it. I hedge on how far down, somewhere between low and high, because I genuinely do not know.
The reasoning earns its price during planning. During execution it usually does not.
Did OpenAI cut the limits in September?
You will find headlines saying allowances were cut by up to 4x for heavy users, and you will find OpenAI saying Astra consumes 3 to 4 times less subscription usage on long-tail workloads. Both claims are circulating in the same week.
I am not going to pick the scarier one. I did not capture the old table, so I cannot show you a before and after, and neither can any of the pages ranking for this query.
What I can show you is the table as it stands today, which is above, and the ratio inside it: Astra is priced against your plan at somewhere between two and two and a bit times GPT-5.6 Sol.
If your allowance feels like it evaporated after the rollout, that ratio is a sufficient explanation on its own. You did not need a secret cut to burn twice as fast. You just needed to start using the new model for everything, which is exactly what everyone did.
I certainly did. My first week I ran maximum effort on questions that did not deserve it.
Resets, including the ones OpenAI handed out
There are three kinds, and they behave differently.
Banked resets. Saved to your account until used or expired. Per OpenAI: "A full banked reset refreshes your five-hour and weekly Work and Codex limits and changes your weekly reset date." You use one from Settings then Usage in the desktop app, where it shows up as something like "1 reset available" or "Full reset."
Automatic or global resets. Applied for you, announced by OpenAI, and they do not appear as a saved reset in Settings.
Purchased instant resets. Available to eligible personal Plus and Pro accounts depending on billing country. They refresh immediately at checkout, cannot be saved, and they pull your whole weekly period forward: your new week starts at your next Work or Codex request, and the next automatic weekly reset is seven days after that.
That last mechanic is worth understanding before you buy one in a panic. You are not adding allowance to this week. You are starting next week early.
And the launch giveaway was real, which is why so many people had spare resets in early September:The launch reset schedule, in OpenAI's own words. Captured September 10, 2026.
Three separate grants: banked resets on September 3 and September 4, and an automatic refresh on September 7 that did not bank anything. OpenAI attributes them to "the broader Astra launch delay," and states that individual access problems after launch do not earn you another one.
Also worth knowing before you get your hopes up: "A reset restores your allowance. It does not permanently increase your plan's limits or change how much allowance a task uses."
What a real meter looks like
Here is my own account today. ChatGPT Pro at $100 a month.My Settings then Usage and billing pane, captured September 10, 2026. Sidebar cropped.
Reading it left to right, because this is what the pages quoting leaked numbers cannot give you:Weekly usage limit: 68% left, resets September 16. That is six days of runway on a meter that read 30% left five days earlier, before the September 7 automatic reset landed.
Credits balance: $0, automatic reload off. No hidden top-up propping this up.
One banked "Full reset" left, expiring October 5. That is what remains of the launch giveaway.On a heavy prompt I can see five to ten percent happen in a single prompt. Not over a session. On one prompt.The same pane on September 5, 2026: 30% left, resetting September 7, the automatic reset OpenAI documents.
I have burned two full resets since Astra shipped. The first one was pure stupidity: I ran the highest effort setting on everything because it was new. The second went to a weekend attempt at building a Pokémon game. It literally burned two usages and I didn't even build the game. It was ugly.
It's only for the hard ones, man, that you should be using Ultra.
I have not hit a wall since. Partly because I had banked resets to fall back on rather than sitting and waiting, and partly because I stopped spending like that.
What to do when you actually hit it
In order, and be honest with yourself at step one:Check which clock stopped you. Settings then Usage tells you. If it is the five-hour window and your weekly meter is healthy, you are not in trouble, you are just early.
Wait, if the job can wait. Five hours is not a crisis. Most jobs can wait five hours.
Move the job down a model, not sideways. Luna gives you roughly 45 to 50 times the messages of Astra. Most of what you were doing did not need Astra.
Spend a banked reset if you have one and the deadline is real. Check the expiry first.
Move the job to another tool. I run Claude and Grok Bot alongside ChatGPT, and when one meter closes I do not sit and stare at it. Grok Bot in particular has been generous with resets.That last point is the real answer to "how do I stop hitting limits," and it is not a limits answer. There is no AI that does it all, and if you are running a business on this you should not rely on one. A single subscription is a single point of failure.
The one-line version
Two clocks, two pools, and Astra costs at least double GPT-5.6 Sol per message. Turn your effort setting down before you turn your spending up.
Comparing the meters across vendors? Claude usage limits explained and Grok usage limits explained are the same breakdown for the other two. If you are deciding between the $100 and $200 ChatGPT tiers, that is ChatGPT Pro $100 vs $200, and what GPT-6 Astra actually is covers the naming. More at Claude Guides.
Published and last reviewed September 10, 2026. Every allowance number is from OpenAI's help article that day, screenshotted above.
The reported September allowance cut and OpenAI's long-tail efficiency claim are both secondary reporting and are labeled as unverified in the text. I did not capture the previous table and cannot show a before and after. The meters are my own paid Pro account with the sidebar cropped, on two dates five days apart. OpenAI's figures are estimated ranges, not entitlements.
Quick answer: GPT-6 Astra is OpenAI's most capable model, released September 3, 2026, with a context window over one million tokens. It is a model, not a new app.
And there is a good chance the word "Astra" does not appear anywhere in your ChatGPT. Your account is fine. It comes down to how OpenAI named things.
I pay $100 a month for ChatGPT Pro. When I opened the model list today to write this page, Astra was not in it.
More on that in a minute, because it turns out to be the most useful thing on this page.
What it is, without the benchmark table
Astra is the model OpenAI now points at its hardest work. From OpenAI's launch post: "state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work," with a context window of more than one million tokens.
OpenAI's own help center is blunter about when to reach for it. It calls Astra "Our most capable model for coding, research, analysis, and complex problem-solving," and the examples it gives are investigating a difficult bug and working through an unfamiliar problem.
Put that list in normal-person words. It does not say "answers your questions better." It says: reads a lot, goes and does things, sticks with a hard problem.
So in one line, Astra is for the jobs that take a while.
For me it is research and a sparring partner. It is good at pushing back on Claude Fable when I want a second opinion on something I already believe. And it is smart enough that the pushback is worth reading.
I have also started using it to build a game, which I will get to.
What I do not open it for is my daily content work. That still goes to Fable, because that is the one that feels like a partner rather than a very good search.
One model, three names
This is what turns a five-minute question into a forty-minute one, and nobody explains it.Where you are
What it is called thereChatGPT Work, and Codex (the developer surface, ignore it if you do not code)
GPT-6 AstraThe ChatGPT Chat tab
GPT-6 ProThe API
gpt-6-astraOpenAI's help center states the split directly: "In Chat, GPT-6 Pro is powered by Astra and is available on eligible Pro, Business and Enterprise plans. Its message limits are separate from the shared Work and Codex allowance and vary by plan. Plus includes Astra in Work and Codex, but not GPT-6 Pro in Chat."
So if you are on Plus, hunting the Chat model picker for the word "Astra," you will hunt forever. It was never going to be there.
Here is my own Chat picker on the $100 Pro plan a few days ago. No Astra anywhere in it.My own Chat picker, captured September 5, 2026, on ChatGPT Pro at $100 a month. The model is in the account. The name is not in the list.
Why you cannot find it, in the order worth checking
Now the live one. This is my Work tab today, September 10, with the model list open.My Work model list on September 10, 2026, on a Pro plan that is entitled to Astra. Default, 5.6 Sol, 5.6 Terra, 5.6 Luna, 5.5, 5.3 Codex Spark. No Astra.
I am not showing you that to claim OpenAI pulled the model. I am showing it because it is the most common thing that happens to real people in week two. OpenAI even documents the fix in its own help center, under a heading called "Update your ChatGPT desktop app."
Their instruction, which is more specific than the usual "try restarting":If Astra is missing, check for updates even if you recently installed one. Your app may need another update and a full restart.OpenAI's help center, captured September 10, 2026. A three-step procedure that exists because this happens to a lot of people.
So here's the checklist, in the order that actually fixes it:Update the desktop app, then fully quit and reopen. Then check for updates again. OpenAI says you may need a second round.
Check which tab you are in. Work and Codex offer Astra by name. Chat offers GPT-6 Pro instead.
Check your plan against the surface. Plus gets Astra in Work and Codex only.My app had crashed to an error screen minutes before I took that shot. When it came back, Astra still was not listed. So updating is a real step, not a brush-off.The state my app was in shortly before the model list above, September 10, 2026. Note which button OpenAI puts first.
When it is there, it looks like this. Same account, five days earlier, running a real job in Work:The same account on September 5, with the Work composer chip reading GPT-6 Astra Ultra. Third name, same model.
Do you have it? The short versionFree: no.
Plus: yes, in Work and Codex. Not in the Chat picker.
Pro $100 and Pro $200: yes, in Work and Codex, plus GPT-6 Pro in Chat.
Business and Enterprise: yes, with seat-level rules.I am deliberately not rebuilding the full surface-by-plan table here, because it already exists: which ChatGPT plans get Astra is the page for that question. This one is about what the thing is.
The part nobody warns you about: it eats your plan
Astra burns your included usage fast. I am on the $100 plan and I have already burned through two full resets.
That is not a vibe. OpenAI publishes estimated messages per five-hour window, and on every plan tier Astra's range is about half of GPT-5.6 Sol's. Same messages, at least twice the burn. The full table and the two clocks behind it are on GPT-6 Astra usage limits explained.
The number worth carrying around: on Plus, the bottom of OpenAI's own estimated range for Astra is five messages in a five-hour window. So think "a couple of messages" for a heavy Astra session.
That's why the setting nobody touches matters more than the model you pick. OpenAI's guidance here surprised me:Lower effort does not mean lower capability across models. For example, Astra at Low effort can outperform Sol at High effort. If you've been getting good results with Sol at High, try Astra at Low or Medium as a starting point.OpenAI's help center, captured September 10, 2026. The most useful paragraph OpenAI published about Astra, and it is buried under a settings heading.
I had heard that claim secondhand and half believed it. It is real, and it is in their documentation.
My own rule ends up in the same place, just from the other side. In my words: use the top setting to plan the hard thing, then drop down to execute it. How far down? Somewhere between low and high. I do not know.
Planning is where the extra reasoning is worth paying for. Doing the work mostly is not. Matching my "top setting" to OpenAI's effort control is my guess, not their instruction.
The full mechanics of the two clocks and three pools are in GPT-6 Astra usage limits explained.
What people are actually building with it
The word-of-mouth is people posting what they made. These are the ones that made me open the model picker. I linked the original posts so you can judge the source yourself.Tom Krcha gave Astra an old drawing of a steam train and asked for it in Blender. His post says it came back as 3,295 editable objects in a few minutes. Original post on X.Roberto Nickson fed it nine phone photos of his studio. It worked out the spatial layout and returned a walkable 3D model. Original post on X.
Three more worth your time, no screenshots because they are threads and video:KP's roundup of the early demos, compiled the day after launch with credit to each builder. The fastest way to see the range.
Min Choi's ten examples: games, 3D scenes, character models, all from prompts or images.
Riley Brown's take: his biggest moment was not visual at all. He gave Astra context on his whole business and let it plan. That matches my own use more than the game clips do.One caveat on all of it. Every one of these people is on a Pro or Business plan or the API. None of them are doing this on the $20 plan, and the next section is why.
The game thing, honestly
The hype about people building games with Astra is real. I have seen shooters. I wanted in, so I spent a weekend trying to build a Pokemon-style game with it.
The graphics it made stunned me. These looked different.Three screens Astra generated for me over the weekend of September 6. Original creatures, original names, a battle system and an overworld. This is what "impressive graphics" meant.
The game did not get built. It burned two full usage resets and what I ended up with was ugly.
I am fairly sure that is me, not the model. I did not ask the computer the right way, and the learning curve is a real curve. But I want that on the record next to the demo videos. The demo videos are all wins. The real week-two experience includes an expensive failure.
If you want to build something serious with it, my read is you want the $200 plan. At $100 you feel the ceiling, and I never felt it with GPT-5.6 Sol.
What I would do this week
If you are on Plus: open the Work tab, not Chat, and give it one hard thing you have been avoiding. A long document you never read, a mess you have not untangled. That is the shape of task Astra is for. Do not burn it on questions.
If you are on Pro: try the thing OpenAI suggests and drop your effort setting a notch. You may find you were paying double for reasoning you did not need.
And if you cannot find it at all, update the app twice before you conclude anything about your plan.
Astra is a real step up on hard, long, multi-step work. Just don't treat it like a better chatbot, or you'll spend a month's allowance on a Tuesday.
Working out which subscription this affects? Which ChatGPT plans get Astra has the surface table, GPT-6 Astra vs Claude Fable 5.1 has the benchmark read, and GPT-6 Astra vs Grok Bot covers the model-versus-teammate question. More of how I run all of this at Claude Guides.
Published September 10, 2026. The effort-level guidance, the Chat-versus-Work split and the update procedure were read and captured that day from OpenAI's help center article "Managing usage with GPT-6 Astra in Work and Codex," and the two screenshots of it above are from that capture. The allowance figures are summarized here and screenshotted in full on the limits post linked above. The launch description and context window come from OpenAI's September 3 announcement, cited as of September 4 when this site first captured it. The account screenshots are my own paid Pro plan, sidebar cropped. My missing-Astra screenshot documents one account on one day and one app build. It is not a claim about OpenAI's rollout.
Quick answer: these two models now cost exactly the same and are good at different things. GPT-6 Astra wins computer use, math, cybersecurity and long context. Claude Fable 5.1 wins the two broadest "how smart is it" boards, including one published by OpenAI itself.
Both are $10 per million input tokens and $50 per million output. So price isn't what decides it anymore.
Every headline yesterday said Astra swept. OpenAI's own launch table does not say that.
The rows nobody quoted
OpenAI published a comparison table on the Astra launch page with Claude Fable 5.1 in a column. Two lines in that table go to Claude.
Every number below is transcribed from that page, checked September 4, 2026. When OpenAI's footnotes put a caveat on a Claude score, I flag it in the next section.Benchmark (OpenAI's own table)
GPT-6 Astra
Claude Fable 5.1Artificial Analysis Intelligence Index v4.1.1
61.2
65.7Humanity's Last Exam (with tools)
57.2%
65.0%Terminal-Bench 4.0
57.9%
55.8%Terminal-Bench Science 0.1
64.6%
52.6%FrontierMath Tier 4 (v2)
97.6%
87.8%GPQA Diamond
96.0%
93.7%AutomationBench
41.4%
31.4%DeepSWE v1.1
74.1%
67.4%FrontierCode 1.1 Main
53.3%
50.9%ARC-AGI-2
95.0%
90.0%Computer use safety, internal (lower is better)
2.4%
9.5%The Artificial Analysis Intelligence Index is about as close as you get to one number for how smart a model is overall. OpenAI put it in its own launch table, and it hands Claude a 4.5 point lead.
Artificial Analysis, quoted independently:Sits beside GPT-5.6 Sol in Intelligence: GPT-6 Astra scores equal to GPT-5.6 Sol in the Index at 61. This is 5 points lower than Claude Fable 5.1 (max with fallback).So on the broadest board, OpenAI's newest model tied its last one.
Read the footnotes before you read the table
OpenAI ran these numbers. That does not make them wrong, but the footnotes change how you should read four rows.BenchCAD: OpenAI's footnote 5 says Claude's scores reflect three modifications to the eval.
ScreenSpot-Pro and ExploitGym: footnote 17 says the Fable scores reported are actually from Mythos, described as "Fable with fewer safeguards." That is a different model.
HealthBench Professional: footnote 11 says OpenAI independently evaluated the Claude models itself, using GPT-5.4 as grader, with Opus 5 substituted when Fable 5.1 refused.
Three science evals: footnote 12 says Claude Fable 5 and 5.1 are excluded from LifeSciBench, GeneBench Pro and MedChemBench because they refuse the majority of questions.None of that is shady. Every vendor benchmarks its own launch. But "state of the art across the board" is how the press summed it up, and the table doesn't say that.
The 99.9% asterisk
The number that traveled furthest was Astra saturating ARC-AGI-3 at 99.9%. It should count the least of anything in the launch.Two problems, both disclosed by the people who ran it:Harness. Per the ARC Prize blog, the 99.9% came from OpenAI's custom "Provider Adapter harness" at about $19K. The default ARC-AGI harness scored 62.7% at about $26K. OpenAI's own footnote 1 confirms it ran a modified responses API harness. That is a 37 point spread depending on plumbing.
No opponent. Claude Fable has no published ARC-AGI-3 result. So the biggest number of the launch has nothing to compare against.The Provider Adapter, in ARC Prize's words, "preserves opaque reasoning state between requests and uses compaction for longer conversations, allowing the model to reuse prior work." That's interesting engineering. But it doesn't tell you how smart the raw model is.
Simon Willison, writing the day it shipped, summed up the launch in one sentence: Astra "appears to score higher than Fable on most of OpenAI's self-reported benchmarks."
The key word there is self-reported.
Where Astra genuinely pulls ahead
Strip out the noise and Astra's wins are real. They just bunch up in a few areas.
Computer use. This is the big new thing in the launch. Astra scores 59.3% on Agents' Last Exam against 55.5% for Claude Opus 5, and OpenAI reports it hitting 72.6% on OSWorld 2.0 in roughly 47% less time per task than GPT-5.6 Sol.
Math and science. FrontierMath Tier 4 at 97.6% against 87.8%. Terminal-Bench Science at 64.6% against 52.6%. These are not close.
Cybersecurity. 100% on ExploitBench, 88.0% single-attempt on SRE-Bench binary reverse engineering. OpenAI says this crosses the Critical threshold in its own Preparedness Framework, so it put limits on the model at launch.
Long context. 100% on OpenAI's eight-needle test at 256K to 512K tokens, and 96.3% at 512K to 1M.
Token efficiency. This one matters more than the scores. On Agents' Last Exam, OpenAI reports Astra using about 65% fewer output tokens than Opus 5 at the highest-scoring settings.
Where Claude Fable 5.1 holds
The broad boards. Intelligence Index and Humanity's Last Exam, both from OpenAI's own table.
Coding, closer than the headline. Astra edges Fable 5.1 on FrontierCode 1.1 Main, 53.3% to 50.9%. But look one column over in OpenAI's table: Fable 5 scores 53.5%, ahead of Astra. On the Artificial Analysis Coding Agent Index, Astra posts 67.0 against 67.2 for Fable 5 and 68.1 for Opus 5. Coding is a wash.
Cost per task at equal quality. Artificial Analysis again: "Per task, the model is less than half the cost of Claude Fable 5, for the same score." That's where Astra's efficiency actually pays off. And it matters more than the trophy rows.
What I actually threw at it on day one
I pay for both stacks, and Astra is one day old as I write this.I didn't run a benchmark. I just ran my actual work, because that's the only test I trust.
The one that changed my mind about Astra was a prospecting task. I asked it to find companies in a specific niche worth reaching out to, and it went and searched forums and threads on its own without me telling it where to look.
What came back was about a month old. That kind of search usually pulls up threads from many months back.
For me that's the real upgrade: it did the digging on its own.Another Astra run from the same week, on my Pro account. I asked it to plan two weeks of paid-work experiments around my real numbers; it pulled live listings on its own and priced the options. The top of the answer has my personal figures in it, so it is trimmed. The chip in the composer is the part worth noticing: on this surface it is labeled GPT-6 Astra Ultra, not GPT-6 Pro.
Then I did the thing I would actually recommend: I ran the output back through Claude Fable to QA it, Fable rewrote the prompt, and I sent it back at Astra. The second pass looked good enough that it's now how I'd run this by default.
Now the other side. I've used Fable more, and for building it still feels more familiar and more dependable to me. But Fable 5.1 has also done some dumb things this week, mostly losing track of tools it definitely had access to. It's minor, but it's real.
My read on the scoreboard: I assumed it said Astra beat Fable outright. It doesn't, and that actually matches what I saw hands-on. Fable still feels better at the deep work. Astra feels newer at getting things done by itself.
The decision rule I would give someone today
Don't switch providers over this. Send each job to the one that's better at it.The job
OpenDeep building, writing in your voice, long careful work
Claude Fable 5.1Anything the AI should do on its own across apps and the web
GPT-6 AstraResearch and volume where you need lots of runs
GPT-6 AstraReviewing and QAing another model's output
Claude Fable 5.1Math, science, anything with a checkable right answer
GPT-6 AstraYou can only pay for one, and you are a generalist
ChatGPT (usage and resets decide it, not the score)I keep coming back to this: smartest isn't the flex anymore. A model that burns tokens to win a board is worth less to me than a leaner one that finishes the task. Both companies are getting pushed that way, and this launch is the clearest sign of it so far.
Verdict
Same price, split by job, and that split is the useful part. Astra is the better agent. Fable is still the better thinker on the broadest boards OpenAI itself published, and it's my daily driver for building.
If you were about to cancel Claude because of a headline, read the table first.
Working out which subscription this actually affects? Which ChatGPT plans get Astra, Claude Max vs ChatGPT Pro, and Claude vs ChatGPT for the everyday call. More of how I run both at Claude Guides.
Published September 4, 2026, one day after GPT-6 Astra shipped. Every benchmark figure was transcribed that day from OpenAI's launch page, and the ARC-AGI-3 harness figures from Simon Willison's write-up citing the ARC Prize blog, both linked above. These are vendor-run numbers on a vendor page and OpenAI's footnotes qualify several Claude scores; read the footnotes section before quoting any row. Benchmarks and prices move fast and this page will need re-checking.
Updated September 22, 2026, the day OpenAI shipped GPT-6 Sol and GPT-6 Luna. First published September 4, when Astra was the only GPT-6 model.
Quick answer: if you pay $20 for ChatGPT Plus, OpenAI says you get all three GPT-6 models: Astra, Sol and Luna. You just do not get any of them in the chat box.
OpenAI puts them inside ChatGPT Work and Codex on Plus, Pro, Business and Enterprise. Edu gets Sol and Luna there too. Chat still runs GPT-5.6 models, plus GPT-6 Pro (that is Astra) on Pro $100, Pro $200, Business and Enterprise.
That is OpenAI's stated rollout, and "rolling out" is the key phrase there. On launch afternoon my own Pro account showed Astra, but not GPT-6 Sol or Luna yet. That was the older app build: an update had downloaded but was not installed.
My rule for picking between them: "You wanna just pay for the optimized model. Don't pay more for less, essentially. That's my model."
Free and Go get one GPT-6 model: Luna, in the desktop app. That part comes from press reports (9to5Mac, VentureBeat), not from OpenAI's help center yet.
Availability is per surface, not per plan. OpenAI has now made that split twice, and it is the whole story. Here is the table.
The plan by surface table, three models wide
Same plan, different answer depending on where you open it. In the Work and Codex columns, Sol and Luna mean the new GPT-6 Sol and GPT-6 Luna.Plan
Chat
ChatGPT Work
Codex
Desktop appFree
GPT-5.6 Luna. No GPT-6
No GPT-6
GPT-5.6 Terra only
GPT-6 Luna (press reports)Go
GPT-5.6 Luna. No GPT-6
No GPT-6
GPT-5.6 Terra only
GPT-6 Luna (press reports)Plus
GPT-5.6 Sol. No GPT-6
Astra, Sol, Luna
Astra, Sol, Luna
Same as Chat and WorkPro $100
GPT-6 Pro (Astra), GPT-5.6 Sol
Astra, Sol, Luna
Astra, Sol, Luna
Same as Chat and WorkPro $200
GPT-6 Pro (Astra), GPT-5.6 Sol
Astra, Sol, Luna
Astra, Sol, Luna
Same as Chat and WorkBusiness
GPT-6 Pro (Astra)
Astra, Sol, Luna
Astra, Sol, Luna
Same as Chat and WorkEnterprise
GPT-6 Pro (Astra), off by default
Astra, Sol, Luna
Astra, Sol, Luna
Same as Chat and WorkEdu
No GPT-6 announced
Sol, Luna
Sol, Luna
Same as Chat and WorkSources, checked September 22, 2026: OpenAI's GPT-6 Sol and Luna announcement as posted to its community forum, the GPT-6 Astra announcement (September 3), and OpenAI's help article on GPT-6 Pro. The Free and Go desktop Luna line comes from 9to5Mac and VentureBeat. OpenAI's help center had not caught up to it when I checked.
For Astra on Plus, the rollout is done. On September 4, OpenAI's Codex and ChatGPT lead Thibault Sottiaux posted on X that "Astra is now rolled out to all Plus and Business users too."
Three details that decide real cases:Rollout is gradual, and OpenAI says availability can differ between Chat, Work and Codex on the same account.
Work and Codex share one usage pool. GPT-6 Pro in Chat has its own, separate limits.
OpenAI has not published message limits for GPT-6 Sol or Luna in ChatGPT yet. Anyone quoting a number is guessing.Launch day on my Pro account: Chat did not move
I checked my own ChatGPT Pro $100 account in the desktop app on September 22, the day Sol and Luna shipped.
One catch up front. The app had an update (version 26.908.70816) downloaded and ready, but not installed. So every launch-day screenshot on this page comes from the older build.
Here is the Chat picker on September 5.And here it is on launch day.Same three rows: Latest, GPT-5.6 Sol, GPT-5.5. No GPT-6 Sol, no GPT-6 Luna. That matches what OpenAI said, since Chat was never on the list.
Then I switched to Work on the same account.GPT-6 Astra is there by name. Next to it sit GPT-5.6 Sol, Terra and Luna. GPT-6 Sol and Luna had not reached my Work list yet when I checked, on that older build. A newer build may already show them.
So "rolling out" means rolling out, even on a $100 plan. If yours is missing too, install any waiting app update and give it a day before you blame your plan.
Same trap, second time
On September 3 OpenAI announced Astra for "all ChatGPT Plus, Pro, Business, and Enterprise users." A day later the help center added the catch:GPT-6 Pro, powered by GPT-6 Astra, is rolling out in ChatGPT for Pro $100, Pro $200, Business and Enterprise plans. Plus plans include GPT-6 Astra in ChatGPT Work and Codex as it rolls out.Most news write-ups quoted the first sentence and stopped. So Plus users opened Chat, saw no Astra, and decided their account was broken.
On September 22 OpenAI did it again, with the surface right in the sentence this time:GPT-6 Sol and Luna roll out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users.The difference: Astra skipped Chat for Plus. Sol and Luna skip Chat for everyone, Pro included.
The naming is not helping. dyauspitr on Hacker News asked it best on launch day: "Wtf is GPT-6 Sol, I though GPT-6 is Astra?"
The number is the generation, the name is the size:GPT-6 Astra is the top model, out since September 3.
GPT-6 Sol is the everyday model for complex work. OpenAI says it is "built for complex coding and agentic workflows."
GPT-6 Luna is the small, fast one, for "high-volume tasks with a clear goal, like summarizing documents, extracting information, or answering quick questions."When to pick which is its own page: GPT-6 Sol and Luna explained.
The picker slot that decides it
Here is the mechanical reason a Plus account cannot show you GPT-6 Pro in Chat.
ChatGPT's picker has reasoning levels: Instant, Medium, High, Extra High, and Pro. GPT-6 Pro lives under the Pro option, next to GPT-5.6 Sol Pro.
Plus does not have the Pro option at all.That table is about GPT-5.6 Sol reasoning levels, but it is the same door. No Pro slot means no GPT-6 Pro, no matter how long you wait. This is a plan gate, not a queue.
Read the number, not the name
If you are scrolling the Chat picker hunting for the word "Astra," you will not find it there.
In Chat it is GPT-6 Pro, per OpenAI's help center. In the API it is gpt-6-astra. In Work, my list now says GPT-6 Astra, and on September 5 the Work composer chip read GPT-6 Astra Ultra.
Sol now has the same problem from the other side. The "GPT-5.6 Sol" in your Chat picker is the older model. GPT-6 Sol is a different, newer model that lives in Work and Codex.
Same name, different generation. Read the number before you decide what you have.
While you are in there: GPT-6 Pro and GPT-5.6 Sol Pro are two different models that share an allowance on some tiers. On the $100 tier they draw from one pool. That is its own decision: ChatGPT Pro $100 vs $200.
What I leave selected on Pro
I pay for ChatGPT Pro at $100. I dropped down from the $200 tier to free up budget for another tool.That is my Usage & billing screen, September 5: Pro at $100 a month, with 30% of the weekly limit left.
So which of the three stays selected? Here is how I put it on launch day:On my Pro account, the hundred dollar one, I probably use Astra on low.In the app, the lowest Astra effort is labeled Light.Next, I'll probably start experimenting with Sol on Ultra, because I think, if I remember correctly, it's cheaper. That's the don't-pay-more-for-less rule at work.
Two honest notes on that. OpenAI has not published how much of your ChatGPT allowance GPT-6 Sol uses, so "cheaper" is not a plan number yet.
It is on the API price sheet, at $2 per million input tokens and $10 output, and inside Work, Astra already costs at least double GPT-5.6 Sol per message (Astra usage limits explained).
And GPT-6 Sol was not in my Work list yet when I checked, on an app build with an update still waiting to install. So the Sol part is my plan, not a result.
Early days versus two weeks in
On September 5 I wrote here that Astra barely moved my usage: a couple of percent across a handful of tasks. That was early days and light runs.
It did not hold. By September 18 I was saying the opposite about Astra:I'd use Sol or just Astra low, but it will burn your usage fast, I mean Astra.Believe the second one. That is exactly why Astra sits on Light and Sol is the next thing I test.
I have also seen OpenAI reset limits generously lately. That is my observation, not an OpenAI commitment, so do not budget around it.
Should you jump from Plus to Pro just to get it in chat?
No. Not for that reason alone. September 22 made the case weaker, not stronger: Pro got nothing new in Chat.
Here is my rule, and it has not changed for any model launch: let the pain tell you. Do not upgrade because a launch made you feel behind. Upgrade when you are actually hitting a wall that costs you time.
Three specific reasons the jump is usually wrong right now:Plus already gets Astra, Sol and Luna. OpenAI's rollout puts all three in Work and Codex. If your work is research, documents or anything agentic, that is where you want them anyway.
The chat allowance is small. Pro $100 is 50 GPT-6 Pro messages a week, shared with GPT-5.6 Sol Pro. That is a real ceiling, not an unlimited upgrade.
$80 a month is a lot to pay for a picker entry. If you want a second frontier model, $20 on a second provider buys a whole separate allowance. That trade is priced out in Claude Pro vs ChatGPT Plus.The honest exception: if you already hit Plus limits every week and wait on resets, you were going to upgrade anyway.
Before you conclude you are missing it
Run this list first. Most "I don't have GPT-6" posts are one of these.Check the surface. Open ChatGPT Work or Codex, not the chat box. That is where all three GPT-6 models live on Plus.
Give it a day. Rollout is gradual. On my own Pro account, GPT-6 Sol and Luna were not in Work on launch day, on an app build with an update not yet installed.
Update Codex. OpenAI says Astra needs Codex CLI 0.153.0 or newer. It has not published a minimum for Sol and Luna yet, so just update to the latest.
Update the desktop app. If an update is sitting there downloaded, install it. On Free and Go, press reports say the desktop app is the only place GPT-6 Luna shows up.
On Enterprise, ask your admin. OpenAI says Enterprise access to Astra is off by default and depends on your workspace's model permissions.If you have done all five on a paid plan and still see nothing, you are in the rollout queue, not excluded.
The one-paragraph version
Per OpenAI's rollout, Plus, Pro, Business and Enterprise get GPT-6 Astra, Sol and Luna in ChatGPT Work and Codex, and Edu gets Sol and Luna there. Chat did not get GPT-6 Sol or Luna on any plan.
In Chat, Astra is called GPT-6 Pro and needs Pro $100, Pro $200, Business or Enterprise. Free and Go get GPT-6 Luna in the desktop app only, per press reports. No plan change is worth making on launch-day panic.
FAQ
Does ChatGPT Plus have GPT-6 Sol?
Yes, in ChatGPT Work and Codex, rolling out from September 22, 2026. Not in Chat: the Sol in the Chat picker is still GPT-5.6 Sol, on Plus and on Pro. OpenAI has not published GPT-6 Sol message limits for ChatGPT yet.
Is GPT-6 Luna free?
Partly. Per OpenAI's rollout as reported by 9to5Mac and VentureBeat, Free and Go users get GPT-6 Luna in the ChatGPT desktop app. OpenAI's help center still lists GPT-5.6 Luna for Free and Go as of September 22, 2026. On the API it is paid: $0.10 per million input tokens and $0.50 per million output.
Is GPT-6 Astra available on ChatGPT Plus?
Yes, but not in the normal chat model picker. OpenAI's help center says Plus includes GPT-6 Astra in ChatGPT Work and Codex.
In the chat picker the model is branded GPT-6 Pro, and that is listed for Pro $100, Pro $200, Business and Enterprise only. Plus is not excluded from Astra, it is looking in the wrong surface.
Why can't I find GPT-6 in my ChatGPT model picker?
Usually one of three things. You are looking in Chat, and GPT-6 Sol and Luna are not in Chat on any plan. The gradual rollout has not reached your account yet. Or you are on Enterprise, where OpenAI says Astra access is off by default and an admin has to enable it.
What is GPT-6 Astra called inside ChatGPT?
GPT-6 Pro, in Chat. OpenAI's help center describes GPT-6 Pro as powered by GPT-6 Astra, and it sits under the Pro option in the reasoning picker alongside GPT-5.6 Sol Pro. In Work it is listed as GPT-6 Astra. Chat has not moved to GPT-6 Sol.
Do Free and Go plans get GPT-6 Astra?
No. Free and Go get no Astra and no Sol. Their Chat default is still GPT-5.6 Luna per OpenAI's help center, and in Codex they get GPT-5.6 Terra only. The one GPT-6 model they do get is Luna, in the desktop app, per press reports.
Does GPT-6 Astra cost extra?
Not by itself. OpenAI says Astra usage is included within existing subscription allowances, and that users will also be able to buy credits for more usage. The allowances are real though: GPT-6 Pro in Chat is metered per week, and Work and Codex share a separate pool.
Do I need to update anything to use GPT-6?
Yes, if you use Codex. OpenAI's help center says Astra needs Codex CLI 0.153.0 or newer and the latest ChatGPT desktop app. It has not published a minimum version for Sol and Luna yet, so update Codex and the desktop app to the latest before you conclude you are missing them.
Correction logSeptember 4, 2026. Published for GPT-6 Astra, the only GPT-6 model at the time.
September 22, 2026. Rebuilt the table for Astra, Sol and Luna after OpenAI shipped Sol and Luna to Work and Codex. Added launch-day screenshots from my Pro account. Qualified my September 5 line that Astra "barely moved my usage" as early days, because two weeks of real use said the opposite.Where this comes from
OpenAI's Sol and Luna announcement, its help center, and the 9to5Mac and VentureBeat reports, all read September 22. The screenshots are my own Pro $100 account, on September 5 and on launch day.
I have not run GPT-6 Sol or Luna yet. They were not on my app build, and an update was still waiting to install.
Deciding between the top tiers instead? That is Claude Max vs ChatGPT Pro. Plan changes as they happen live on the AI plan tracker, and everything I run and pay for lives on Claude Guides.
Chris Alarcon 
