ChatGPT Usage Limits (2026): What Plus, Pro, and Business Actually Get, and the 5-Hour Window

ChatGPT usage limits: the short answer

As of September 2026, everyday text chat in ChatGPT is unlimited on every plan, Free included, subject to abuse guardrails. The limits that actually run out are three separate meters: reasoning with GPT-5.6 Sol, GPT-6 Pro messages in Chat, and a shared ChatGPT Work and Codex allowance measured in rolling 5-hour windows. A ChatGPT Business Standard seat gets Plus-level Work and Codex usage; a Premium seat gets 5x that with no 5-hour limit. Every number below comes from OpenAI's help center, pricing page, and Codex pricing docs, read on September 12, 2026.

Most "ChatGPT limits" articles still quote a message count per three hours from a model that no longer exists. OpenAI rebuilt the whole meter this summer. Here's how it works now.

What changes in December 2026: one shared usage pool

Starting in December 2026, Plus, Pro and Business plans get one usage allowance shared across ChatGPT and Codex, measured in tokens instead of message counts. That's from OpenAI's help article Understanding ChatGPT and Codex usage limits (read November 9, 2026). OpenAI hasn't given an exact day, just "December".

• What counts: your ChatGPT work and your Codex work draw from the same allowance. Bigger models and harder tasks use more of it.
• What doesn't count: Instant chat. OpenAI says it "will not draw from your usage."
• Where to check: Settings, then Usage. That screen shows your combined remaining usage and when it resets.
• When you run out: wait for the reset, upgrade your plan, or buy credits.

What it means in practice: if you lean on Codex from a Plus seat, that coding now spends the same budget as your long ChatGPT work sessions. OpenAI says it expects most people won't wait on resets more often than they do now. I'd still check Settings, then Usage, in the first week of December before paying for an upgrade. The per-plan numbers below describe how limits work today.

The three meters that actually run out

1. Reasoning. Choosing Medium, High, or Extra High uses GPT-5.6 Sol. When you hit that limit, OpenAI says ChatGPT "may continue with another available reasoning model" and shows the reset time when it can.
2. GPT-6 Pro in Chat. OpenAI's top Chat option, powered by GPT-6 Astra, has hard allowances by plan. "A Pro subscription does not include unlimited use of the GPT-6 Pro model."
3. ChatGPT Work and Codex. These share one agentic allowance, metered like Codex: an estimate per 5-hour window, and "weekly limits may also apply." OpenAI's rate card says ChatGPT for Excel and Workspace Agents draw from the same pool.

Separately, tools like file uploads, image generation, voice, and data analysis carry their own limits on the lower plans. Plain chat is the one thing you won't run out of.

ChatGPT usage limits by plan

Two footnotes that matter. On Pro $200, GPT-5.6 Sol Pro has its own daily allowance plus a combined daily cap with GPT-6 Pro; OpenAI published those counts on September 12 and no longer does. And the Pro $200 pause is over, along with the cut that came with it: on October 29, OpenAI "canceled the Pro 200 reduction announced in our September 29 email," so subscribers on the lower allowance got twice their previous Work and Codex allowance. Pro 100 and Pro 500 doubled the same day. There's also a new Pro 500 at $500 a month, the only Pro tier with Astra Ultrafast. My verdicts on those tiers are on is ChatGPT Plus worth it and is ChatGPT Pro worth it.

OpenAI doesn't publish a message count for Plus reasoning, so I'm not going to invent one. It shows the reset time in the product when you hit it.

ChatGPT free limits per day: how many messages do you get?

There is no published per-day number, because OpenAI doesn't meter ChatGPT by the day. On the pricing page (read September 12, 2026), everyday text chat is listed as unlimited on Free, Go, Plus, and Pro, "subject to abuse guardrails." So a Free user who only chats doesn't run into a daily count.

What a Free user does run into is the heavier work, and those meters run on rolling five-hour windows rather than calendar days, so "per day" is the wrong clock. ChatGPT Work is listed as limited, desktop app only, on Free and Go. Codex on Free and Go is GPT-6 Luna at Standard speed in the desktop app, subject to rollout, and image generation inside Codex isn't available on Free at all (OpenAI's Codex pricing page, read October 10, 2026). When a window runs out, you wait for the reset time ChatGPT shows you, which is usually hours, not until tomorrow.

If you keep hitting those walls, the plan-by-plan table above is the honest comparison: Go at $8 a month raises Free's ceilings, and Plus at $20 adds Codex across the web, CLI, and IDE with an estimated 15 to 160 GPT-6.1 Sol messages per five hours.

If you're choosing a plan rather than managing one, every ChatGPT price is on one page: Free $0, Go $8, Plus $20, Pro $100, $200 or $500, and Business $20 a seat billed yearly or $25 monthly, with what each plan runs out of. That's in ChatGPT pricing.

The 5-hour window, explained

ChatGPT Work and Codex don't count messages the way chat used to. They count how much work each message causes. OpenAI's estimates, in local messages per five hours:

• Plus and Business Standard: GPT-6 Astra 5 to 45, GPT-6.1 Sol 15 to 160, GPT-6 Sol 15 to 150, GPT-6 Luna 350 to 3,000. OpenAI's Business article still lists GPT-5.6 Sol 10 to 100, Terra 25 to 200 and Luna 250 to 2,000 for Standard seats (and puts GPT-6.1 Sol at 15 to 150); the Codex page dropped the GPT-5.6 rows.
• Business Premium: 5x Standard's included usage, with no 5-hour limit.
• Pro ($100, $200, $500): "Pro plans currently have no five-hour limit." Weekly limits can still apply.

The ranges are wide on purpose. OpenAI says a longer task "can use more of your allowance than a short task, so a fixed number of messages is not a reliable measure of remaining usage." Cloud runs can burn more than local ones. GPT-6 Astra, the model everyone reaches for first, burns fastest. The detailed Codex side, including how credits map to tokens, is on Codex usage limits.

The practical read: if you hand Astra a vague "clean up this whole folder" job at 10am, you can be locked out of Work until mid-afternoon. Point Terra or Luna at the grunt work and save Astra for the job that needs it. (The GPT-6 Sol and GPT-6 Luna rows are new since this page was first written: they are on the September 28 read of OpenAI's Codex page, not the September 12 one. GPT-6.1 Sol and Pro's missing five-hour limit arrived on the September 29 read.)

ChatGPT isn't the only assistant on a five-hour clock. Claude resets on one and Google refreshes Gemini on one; what each vendor meters inside its window, and which ones print a number at all, is laid out in the AI usage limits by plan table.

Hitting the Plus window every week is the most common reason to move up a tier. Whether Pro 100 is worth $80 more for you is worked out on ChatGPT Plus vs Pro.

ChatGPT Business limits: Standard vs Premium seats

For a company, the seat type is the limit decision. OpenAI's Business help article puts it plainly: "Premium seats have 5x the included usage of Standard seats, with the benefit of no 5-hour limit."

Standard seats include "limited Astra usage" inside their Work and Codex allowance. Premium seats can spend their full allowance on Astra. In Chat, Standard gets 15 GPT-6 Pro messages a month and Premium gets 50 a week. Everyday Instant chat is unlimited on both.

You can mix seat types in one workspace and move people between them. My rule for small teams: Standard for everyone, Premium only for the one or two people who hit the wall in the first month. The dollars behind that choice are on ChatGPT Business pricing.

What is the agentic usage limit in ChatGPT?

It's the meter for work ChatGPT does on its own, as opposed to answering a chat message. ChatGPT Work and Codex draw from it, and OpenAI's pricing docs (read October 14, 2026) say other agentic features join the same pool once their pricing takes effect; today that includes ChatGPT for Excel on Plus and Pro. Your everyday chat messages are a separate meter, which is why someone can be "out of usage" in Work while regular chat still answers.

On Plus and a Standard Business seat, OpenAI publishes the same estimates per five-hour window, counted in local messages:

• GPT-6.1 Sol: 15 to 160
• GPT-6 Sol: 15 to 150
• GPT-6 Astra: 5 to 45
• GPT-6 Luna: 350 to 3,000

Weekly limits may apply on top, cloud tasks use more than local messages, and Pro currently has no five-hour limit. The range is wide because a message that reads a big folder, calls tools and reasons for a while costs far more than a short one. OpenAI says prompt length alone isn't a reliable estimate.

The setting that quietly halves it is speed. Fast mode spends included usage at 2.5x the Standard rate, and Ultrafast at 8x. Run GPT-6.1 Sol in Fast mode and the 15 to 160 band behaves like roughly 6 to 64 messages; in Ultrafast, about 2 to 20. If a teammate keeps hitting the agentic limit by lunch, check their speed setting before you buy a Premium seat.

When you hit it mid-task, OpenAI says the turn in progress gets to finish. After that you wait for the reset, or on Plus and Pro you can buy credits instead of upgrading.

Out of Work usage? What each limit does when you hit it

• Reasoning limit: ChatGPT may switch you to another reasoning model until the reset.
• Plus and Pro: you can buy additional credits instead of upgrading.
• Business: eligible usage continues on workspace credits if the workspace has them and spend controls allow it. Owners buy credits under Workspace settings, then Billing. Purchased credits are valid for 12 months and can reload automatically.
• Nobody: gets a reset from Support. OpenAI says Support "does not reset ChatGPT or Codex usage limits."

For budgeting, OpenAI's rate card lists approximate credit costs: 10 credits for a GPT-5.6 Sol message, 50 for a GPT-6 Pro message, 50 for a deep research task, 5 for an image, and 1.25 per minute of voice. Instant chat doesn't use credits.

Put those on one scale and the card becomes an exchange rate: one GPT-6 Pro message or one deep research task costs the same as 5 Sol messages, 10 images, or 40 minutes of voice. That's the comparison to make before someone reaches for Pro out of habit.

A worked week, to size a credit cap. Take one analyst on a Business seat who has already used the seat's included allowance and keeps going on credits for a five-day week, running 3 deep research tasks, 30 Sol messages and 10 images a day:

The workload is my assumption, and the card's numbers are approximate. OpenAI doesn't publish a dollar price per credit on the pages I read (it depends on your plan or agreement), so the budget you set is in credits, not dollars. Swap two of those deep research tasks for Sol messages and the day drops to 420; swap them for GPT-6 Pro messages and nothing changes, because both cost 50.

One setting to change on day one: OpenAI says that by default, "all seats and users have no limits specified" for credit usage. Set a monthly credit cap per seat type before someone's weekend experiment with Astra finds it for you.

How many images can you make on each ChatGPT plan?

Short answer: OpenAI doesn't print a number for any personal plan, and I'm not going to make one up. Here's what its pricing page actually commits to, read November 5, 2026, the day after Images 2.6 shipped:

Pro is the only plan where OpenAI uses the word unlimited for images. Plus doesn't promise more pictures than Go, it promises better ones: the Thinking version that plans the layout before it draws. That's the one you want for anything with words in it, like a flyer or a menu board. When a lower plan runs out, ChatGPT tells you when the image limit resets. It's a separate meter from chat, so you can keep talking while you wait.

Business and Enterprise workspaces are the only places with a printed price. OpenAI's rate card charges about 5 credits per image generation once a seat is past its included usage. Edu workspaces get 3 image generations per rolling 24 hours that don't touch credits at all. On the credit scale above, a marketing person making 40 social images a week past the allowance spends 200 credits. That's the same as four deep research tasks. Fine for one person, and worth a per-seat cap before the whole team starts doing it.

What changed on November 4: Images 2.6 is faster and better at keeping details the same across edits (OpenAI's example is a character's outfit staying put when you swap the background). ChatGPT can now make short looping GIFs, and you can save a reference photo, like your product or your storefront, and call it by name in later chats instead of uploading it again. OpenAI didn't announce any limit change with it. If you need volume more than convenience, the API bills per token instead. Same model, no plan meter, and the per-million rates are on OpenAI API pricing.

How I'd keep a team from hitting the wall

1. Match the model to the job. Terra and Luna for sorting, extracting, and first drafts. Astra for the deck the client actually sees.
2. Scope Work tasks tightly. "Use only these three files, compare the top five options, stop before sending anything" costs far less than "figure it out."
3. Give Premium to the heavy users, not the org chart. Usage analytics after 30 days will tell you who they are.
4. Cap credits before you need to. A monthly limit per seat type is the difference between a known bill and a surprise.

If the thing burning your allowance is an actual multi-step job, it's worth setting up properly instead of re-prompting it daily. That's what ChatGPT Work is for, and its limits are the same meter covered above.

If your team re-pastes the same instructions and files into every chat, put them in a shared Project instead (up to 40 files on Business). And if you built custom GPTs for that job, OpenAI plans to retire them: on its published Enterprise timeline, GPTs stop running on December 11, 2026. Which one to use now is in ChatGPT Projects vs custom GPTs.

If you use ChatGPT's models through DuckDuckGo's private Duck.ai instead, its limits work differently: a daily budget that resets at midnight UTC and, on paid tiers, a weekly one on top. That's covered on Duck.ai usage limits.

Sources

• OpenAI Help, GPT-5.6 and GPT-6 Pro in ChatGPT (reasoning fallback, GPT-6 Pro allowances by plan): read September 12 and 29, 2026
• ChatGPT pricing (unlimited everyday chat, abuse guardrails, Work on Free and Go): read September 12, 2026
• OpenAI, Codex pricing (now served at learn.chatgpt.com/docs/pricing; Work and Codex estimates per five hours, Pro's missing five-hour limit, weekly limits, credits): read September 12, 28 and 29, 2026
• OpenAI Help, ChatGPT Business models and limits (Standard vs Premium seats, Business GPT-6 Pro allowances): read September 12 and 29, 2026
• OpenAI Help, ChatGPT Business credits (12-month validity, auto-reload, per-seat credit limits): read September 12, 2026
• OpenAI Help, credit rate card (credits per message, task, image and minute): read September 12, 2026
• OpenAI Help, ChatGPT release notes (Images 2.6, GIF creation, saved reference images, November 4) and the image rows of ChatGPT pricing and the credit rate card (5 credits per image; Edu 3 free per 24 hours): read November 5, 2026
• OpenAI Help, About ChatGPT Pro tiers (Pro 100, 200 and 500; Pro 200 reopened; reduction canceled): read September 12, September 29 and November 1, 2026