GPT-6 Astra limits: what Plus and Pro actually get

ChatGPT Plus does get GPT-6 Astra, but only inside ChatGPT Work and Codex, at roughly 5 to 45 messages per rolling five-hour window. Astra in the ordinary chat box needs Pro, Business or Enterprise, where it appears as GPT-6 Pro on a weekly allowance of 50 or 200 messages. Free and Go get no Astra at all.

The short answer

The part nobody expects: Astra is not in the Plus chat box

OpenAI launched Astra on 3 September 2026, describing it as a new frontier on computer and browser use and rolling it out to Pro, Plus, Business and Enterprise accounts over the following week, as TechCrunch reported at launch. Read that sentence quickly and you conclude that a $20 Plus subscription now puts the best model in the model picker.

It does not. The model picker in regular Chat still tops out at GPT-5.6 Sol for Plus. Astra arrives somewhere else entirely: inside ChatGPT Work and inside Codex, the two surfaces OpenAI built for longer autonomous jobs. Codex additionally needs CLI version 0.153.0 or newer before Astra shows up at all.

This catches people out because every previous model launch worked the other way round. GPT-5 and GPT-5.6 appeared in the chat box first and in the tooling later. Astra inverted it, and enough Plus subscribers noticed that the change became a standing feature request on OpenAI's own community forum. If you are on Plus and cannot find Astra, nothing is broken and no rollout is pending. You are looking in the wrong place.

GPT-6 Astra limits by plan

OpenAI publishes estimated ranges rather than message counts, because consumption depends on the work. Here is the full picture as of today, drawn from OpenAI's help pages as reported by Engadget and Notebookcheck.

Three things are worth pulling out of that table before moving on.

First, the Chat allowance and the Work allowance are separate budgets. Spending all 50 weekly GPT-6 Pro messages on the $100 tier does not touch your five-hour Codex window, and emptying the Codex window does not cost you a Chat message. They are two meters, not one.

Second, on the $100 Pro tier the weekly 50 is shared with GPT-5.6 Sol Pro. Every Sol Pro answer is one fewer Astra answer that week. That detail turns "50 a week" into something materially smaller for anyone who used Sol Pro habitually before Astra existed.

Third, the Plus range is not a typo. 5 to 45 is a nine-fold spread, and the reason for it is the next section.

Why Astra empties a window so much faster than Sol did

The meter measures compute, not messages. That has always been true of ChatGPT's usage windows, but Astra makes the difference impossible to ignore for three compounding reasons.

1. The model itself is more expensive per token

Astra's API rates sit several times above GPT-5.6 Sol's. Whatever conversion OpenAI applies between subscription allowance and underlying compute, a token answered by Astra is worth more of your window than the same token answered by Sol.

2. Reasoning effort multiplies it again

Astra exposes a reasoning-effort setting, and higher effort means more internal tokens for the same visible answer. Two users sending the identical prompt can consume very different amounts of window depending on that one control.

3. Computer use turns one message into a long run

This is the big one, and it is the reason OpenAI's own numbers are a range. Astra's headline capability is operating a computer and a browser. A single instruction like "go through this repository and fix the failing tests" is one message by the counter and potentially hundreds of tool calls by the meter. Fortune reported that this is precisely what strained OpenAI's capacity after launch, because Astra interacts with desktops at superhuman speed and burns through allowances faster than previous models.

We wanted to take the smallest step that allows us to continue giving the broadest access possible. Thibault Sottiaux, OpenAI product lead, on pausing $200 Pro sign-ups

Put the three together and the practical shape is this: a Plus subscriber doing short, well-scoped Codex edits lands near the top of the 5 to 45 range. A Plus subscriber pointing Astra at a large refactor on high reasoning effort can land at the bottom of it, and be locked out of Astra twenty minutes into a five-hour window. Neither person is doing anything wrong. They are buying different amounts of compute with the same subscription. We cover the wider mechanics in our guide to how ChatGPT's limits really work in 2026.

Two clocks, and you need room in both

Where a plan applies both a five-hour and a weekly limit, allowance has to remain in both for you to keep going. This is the mechanic that produces the most confused support posts, so it is worth stating plainly:

• Five-hour window empty, week healthy. You stop until the window rolls over, even though your weekly number looks fine.
• Week empty, fresh five-hour window. You stop anyway. A new window does not refill a spent week.
• Both have room. You continue, and both counters go down together.

The five-hour window is rolling rather than fixed to the clock. A new window begins when you send your first message in Work or Codex after the previous one has ended, which means your reset time drifts with your own behaviour rather than sitting at a predictable hour. That is the same trap ChatGPT's older rolling caps set, and we walk through the arithmetic in when does ChatGPT reset.

The $200 Pro tier: paused, then reopened with half the allowance

The most consequential change since launch has nothing to do with the model and everything to do with what you can buy.

On 10 September 2026 OpenAI paused new sign-ups and upgrades to the $200 Pro tier, citing capacity strain from Astra demand. Existing $200 subscribers kept their plan; the $100 tier and Plus stayed open. Sottiaux called the demand "unprecedented", adding that he had "not seen anything like it until now and we went through very steep growth before".

On 29 September 2026 OpenAI announced the tier would reopen the following day, with a recalculated allowance. Per Unite.AI and Android Authority, the new calculation nets out at roughly half the API-spend value of the old $200 plan. OpenAI's argument is that two other changes offset the cut:

• Cheaper models. GPT-6 Sol and GPT-6 Luna arrived on 22 September 2026 at around half their predecessors' API pricing, so each dollar of allowance stretches further on everything that is not Astra.
• No five-hour window on that tier. OpenAI committed to not reintroducing the five-hour cap for $200 Pro, so the weekly allowance can be spent whenever you want rather than rationed into five-hour slices.

Whether that nets out in your favour depends entirely on your shape of use. If you were concentrating heavy Astra work into bursts and hitting the five-hour wall, removing the wall is worth real money. If you were spreading work evenly and simply consuming a lot of it, halving the allowance is a cut with a nicer explanation attached. Our honest read on whether ChatGPT Pro is worth it works through the $100 against $200 decision, and the same question for Plus covers the tier below.

Free and Go: no Astra, and that is less of a problem than it sounds

Neither Free nor Go includes Astra in any surface. Both default to GPT-5.6 Luna, and ordinary text chat on both has been uncapped since OpenAI removed message caps from text chats in August 2026. Go, at $8 a month, buys context window, uploads, image generation and memory rather than a better model.

For a large share of everyday use that is a fair trade. Astra's advantage is autonomous, tool-heavy, long-running work. If you are asking questions, drafting, summarising and coding interactively, Luna and Sol do that without a meter running at all. The moment to pay for Astra is when you want to hand over a job and walk away, and that is exactly the moment the meter matters most.

What to do when the Astra window runs out

It is not a lockout of ChatGPT. It is a lockout of one model in one surface. Four options, roughly in order of how often they are the right one:

1. Finish on Sol or Luna. Regular text chat is uncapped. Most of what remains in a half-finished task is explanation, review and small edits, and the cheaper models do that perfectly well.
2. Wait out the rolling window. Because the window is rolling, the wait is at most five hours from your first message in the current window, often much less. Our what to do when you hit a ChatGPT limit guide covers how to use that gap productively.
3. Buy credits. Plus and Business Standard seats can top up with optional credits once the included Astra allowance is gone. Worth it for a deadline, expensive as a habit.
4. Move the work to the other provider. If you also pay for Claude, a spent Astra window is a good reason to run the next task there instead. We compare the two limit systems side by side in Claude vs ChatGPT limits.

How to see where you actually stand

ChatGPT will tell you that you have hit a limit. What it does not do is tell you, before you start, that the job you are about to launch will not fit in what is left. On a model whose whole point is that you set it going and stop watching, that gap is the entire problem.

Checking manually works if you remember to do it, and nobody remembers to do it at the only moment it would have helped. A passive meter closes that gap: a remaining percentage and a reset countdown sitting in the menu bar turn the window from something you discover into something you plan around. That is all tokn.watch does. Connecting takes about a minute, and the setup steps are on our homepage. If you want the full range of options first, including the ones that are not ours, our roundup of ways to track ChatGPT usage lays them out.

The verdict

Astra's limits are not really about numbers. The numbers are published, they are reasonable for the compute involved, and they are in the table above. What makes them feel unfair is the mismatch between how the model is sold and how it is metered: a model built to work autonomously for long stretches, sold on plans metered by a window you cannot see, in surfaces most subscribers did not know existed.

Three things to take away. Plus gets Astra in Work and Codex, not in Chat. The five-hour and weekly clocks are separate and both have to have room. And the published range is wide because the cost of a message is set by what you ask it to do, not by the fact that you sent one. Know which of those three is biting before you conclude that you need the next tier up.

Frequently asked questions