Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
ChatGPT limits in 2026: every number OpenAI publishes, plan by plan
Every ChatGPT limit OpenAI publishes, plan by plan: chats, thinking, context windows, files, projects, voice and storage, and what it keeps quiet.
Prashant Shaw
Founder, Sumibako
8 min read
Ask five websites what ChatGPT Plus's limit is and you get six answers, most of them describing a ChatGPT that no longer exists. In the last six weeks OpenAI launched GPT-6, split Pro into three prices, retired automatic switching to Thinking, and rewrote most of its help pages, several of them within the last day.
So this post prints only numbers from OpenAI's own pages, says plainly where OpenAI prints none, and points out the two places where its pages disagree with each other. Everything was checked on 22 October 2026, and the links go to the pages themselves, so you can check what changed since.
On every plan, the pricing page lists everyday text chats as unlimited. The Free plan FAQ says: "Free users have unlimited everyday text chats, subject to abuse-prevention safeguards. File uploads, image generation, voice, data analysis, and other tools have separate usage limits."
So the limits people hit in 2026 are not on chatting. They are on three other things.
Thinking. Free and Go get Think, which the models page says "uses GPT-5.6 Luna". Plus gets the Medium and High thinking levels, but not Extra High or Pro. Pro gets all four. OpenAI does not publish how many thinking messages Plus gets, only what happens at the end: "If you reach a Thinking limit, ChatGPT may continue with another available model."
No more automatic thinking. Since 14 September, per the release notes: "We’re retiring automatic switching from Instant to Thinking (reasoning) for ChatGPT Plus and Pro users globally." The models page adds that asking it to think harder in your message no longer switches levels either. If you are on Plus and answers got worse in September, check the slider before you check your prompts.
Pro models. The models page says: "A Pro subscription does not include unlimited use of the GPT-6 Pro model." On Pro $200, when you reach the weekly GPT-6 Pro limit, ChatGPT drops to GPT-5.6 Thinking at Medium by itself. On Pro $100, GPT-6 Pro and GPT-5.6 Sol Pro "draw from one weekly allowance". Neither weekly number is published.
🧠 How much a conversation can hold
The context window is how much text ChatGPT can consider at once: your messages, its replies, files and everything else in the chat. From the pricing page:
Free
Go
Plus
Pro
Instant, total window
27K tokens
54K
54K
128K
Instant, most you can paste in
about 12 pages
about 40 pages
about 40 pages
about 250 pages
Reasoning, total window
Varies
256K
256K
400K
Reasoning, most you can paste in
Varies
about 320 pages
about 320 pages
about 680 pages
The pages column is smaller than the window because, in the pricing page's footnote, "The portion available for user input is smaller than the total window, as space is also used for system instructions (including tools and personality), memories (if enabled), and internal processing". Your saved memories ride along in every chat and take up room.
What OpenAI does not publish is how long one conversation can run before ChatGPT tells you it has reached its maximum length. That wall and the window are related but not the same, and I wrote up what is known about the conversation limit separately. If your message said "Chat memory full" instead, that is a different post.
How many: "You can upload up to 80 files every 3 hours. Free users are limited to 3 file uploads per day. OpenAI may lower these limits during peak hours."
How big: 512 MB per document or presentation, and "Text and document files are capped at 2 million tokens per file." Spreadsheets about 50 MB, images 20 MB, audio 512 MB.
Where you stand: "ChatGPT does not currently show how much of your rolling upload-rate quota has been used or remains." You find out by hitting it.
Projects: 5 files per project on Free, 25 on Go and Plus, 40 on Pro. The projects page puts collaborators at 5, 10 and 100, and project sharing is on every plan.
Custom GPTs: up to 20 knowledge files, for as long as GPTs last, which is until 11 December.
Storage is where OpenAI's pages disagree for the first time. The Library page gives 500 MB on Free, 4 GB on Go, 20 GB on Plus and "Pro users have up to 100 GB of Library storage." The uploads page says "Shared upload storage is capped at 25 GB per user". Both can be true only if they measure different things, and neither page says so. On Pro, I would plan around the smaller number until they agree.
🎙️ Voice
From ChatGPT Voice, measured "over a rolling 24-hour period":
Plan
Live voice
Free
Limited, on GPT-Live-1.1 Quip. "Limits may change."
Go
3 hours, on GPT-Live-1.1 Quip
Plus
3 hours
Pro $100
15 hours
Pro $200
Unlimited, "subject to abuse guardrails"
The page does not list Pro $500. Presumably it is at least as generous as Pro $200, but presumably is not a number.
💸 The three Pro plans
The Pro tiers page lists Pro 100, Pro 200 and Pro 500, and says only that each includes more usage than the one before. What the extra money buys that you can measure:
Ultrafast, on Pro 500 only. It spends your included Work and Codex allowance "at 8× the Standard rate for the selected model", then credits at 6 times. Fast mode spends it at 2.5 times.
Voice: 15 hours on Pro 100, unlimited on Pro 200.
If you had Pro 200 at any point from 22 to 29 September: you keep the old, larger allowance "through October 29, 2026", then move to the new one at the same price.
Work and Codex have their own allowances, separate from chat. The Work and Codex page warns that "Astra may use your allowance faster than GPT-5.6 Sol", which is worth knowing before you set it as your default.
🙈 What OpenAI does not publish
None of these has a number on any OpenAI page I could find. The pricing page says "Limited", "Expanded" or "Maximum" instead:
How many thinking messages Plus gets, and the weekly GPT-6 Pro limits on Pro.
How many images you can make. Pro's is "Unlimited and faster image creation"; the rest have no figure.
How many deep research tasks you get. The deep research page says only that a counter in the app shows what is left, and that a monthly allowance "resets every 30 days from the date of your first use."
How much memory holds. The pricing table says "Limited" for Free, "Yes" for Go and "Expanded" for Plus and Pro, and the Pro card says "Maximum memory and context". None of them is a number.
How long a conversation can get.
And the second disagreement: which model Free and Go use. The release notes of 7 October say "Free and Go use GPT-6 Luna", and the pricing page lists GPT-6 Luna on Free. The models page, updated since, says "Free and Go use GPT-5.6 Luna" for Instant. If you are on Free and want to know, the model picker on your own account outranks all three.
NOTE: Guides that print an exact number of messages for Plus are not quoting any current OpenAI page I could find. They may be quoting an older page, or estimating, and either is fair, but neither is what OpenAI says today.
Anthropic is no more forthcoming: Claude's usage limits come as multiples of a five-hour session, not as counts.
🔁 When you hit one
1️⃣ Read the message. Per OpenAI's Pro tiers page, ChatGPT "displays the reset time when available", and a usage limit "does not end your subscription or mean your account is restricted."
2️⃣ Switch model. Most limits are per model or per tool. Plain text chat keeps working.
3️⃣ Do not ask Support to reset it. In the words of the same page: "Support cannot reset ChatGPT or Codex limits." It can look into usage that was counted wrongly.
4️⃣ Know what upgrading does. On Free, the Free plan FAQ says "your usage rates will reset when you update your account to a Plus plan, a Pro plan, or a Business plan." Buying credits on Pro is different: per the Pro tiers page, "Credits do not provide additional usage for every ChatGPT model or feature", only for eligible ones such as Codex and Work.
If the limit you keep hitting is the end of a long conversation rather than any of these, the cure is a handoff note kept outside the chat, which is what the length post is about, along with the notes app I build for keeping it.
ChatGPT's limits are like a gym's fair use policy: unlimited, until it is not, and the sign with the actual number is in the back office. At least now you know which doors to try.
Eight eMudhra keys were integers under 256, one of them 4. Why browsers blocked them in hours, why your Docker image did not, and what to check by 27 October.
