In 60 seconds: ChatGPT's everyday chat is unlimited on every plan — including free — but that sentence hides the real story. The limits that actually bite are on reasoning models, GPT-6 Pro chats, and the Work/Codex agent allowance, measured in 5-hour windows, weekly caps, and a 1.05-million-token context window. Here's the full tier-by-tier breakdown, checked against OpenAI's own pages today.
- Everyday text chat is unlimited on Free, Go, Plus, and Pro — abuse guardrails aside — per OpenAI's pricing page.
- The real limits sit elsewhere: reasoning-model allowances, GPT-6 Pro chat caps (weekly), and a shared Work/Codex allowance in 5-hour windows plus weekly caps.
- Context windows are 1.05M tokens on GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, with 128K max output — per OpenAI's official model docs.
- The API is a different world: pay per token, no plan caps — rate limits (RPM/TPM) are the constraint there.
- One million tokens is roughly 750,000 English words — but your usable share of the window is smaller than the headline number.
Nobody Googles "token limits" for fun. You Google it at 11:47 p.m., mid-project, when ChatGPT tells you you've hit a limit — and the message doesn't say which limit, when it resets, or what you did to deserve it.
Here's the thing: most "ChatGPT limits" articles quote numbers from a metering system OpenAI already replaced. Every figure below was checked against OpenAI's own pricing page, model docs, and help center on October 21, 2026.
Visualizing a context window: tokens filling the meter — AI-generated illustration
What happened
OpenAI rebuilt how ChatGPT is metered this year. The old world — "40 messages every 3 hours" — is gone. Now plain everyday chat is unlimited on every plan, Free included, subject to abuse guardrails. Genuinely good news.
The limits didn't disappear, though. They moved — onto the expensive stuff: reasoning models, the top-tier GPT-6 Pro chat option, and the agentic Work/Codex allowance. Never hit a limit? You're living entirely in everyday chat. The moment you ask the model to think hard or hand it a long agent task, you enter metered territory.
Latest verified facts
First, the plan ladder, as listed on chatgpt.com/pricing:
| Plan | Price | Chat model | Everyday chat | What actually runs out |
|---|---|---|---|---|
| Free | $0 | GPT-6 | Unlimited text chats | Uploads, image creation, voice, deep research, Codex/Work access — all limited |
| Go | $8/mo | GPT-6 | Unlimited text chats | Higher tool, upload, image, and voice ceilings than Free; plan may include ads |
| Plus | $20/mo | GPT-6 + advanced reasoning | Unlimited text chats | Reasoning-model allowance; GPT-6 Pro chat capped; Work/Codex in 5-hour windows + weekly cap |
| Pro | $100 / $200 / $500/mo | GPT-6 Astra pro reasoning | Unlimited text chats | Weekly Work/Codex allowance (5x / 10x / 25x of Plus); GPT-6 Pro weekly prompt caps; $500 tier adds Ultrafast speed |
Now the tokens. Per OpenAI's official model docs, the current flagships share the same headline numbers: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna each offer a 1.05-million-token context window with 128K max output tokens. A token is roughly three-quarters of an English word — so 1.05M tokens is ~750,000+ words, dozens of books, in one conversation's memory.
But OpenAI states plainly in its own pricing-page footnote: ChatGPT manages a shared context window, and your input gets less than the total. System instructions, tools, memories, and the model's own reasoning all drink from the same 1.05M-token pool. A long chat with memory and tools on can feel "full" long before you've typed 750,000 words yourself.
On the metered side: OpenAI's help center states a Pro subscription does not include unlimited use of the GPT-6 Pro model — its chat allowance runs on weekly prompt caps per Pro tier. The Work/Codex allowance is shared, measured in rolling 5-hour windows plus a weekly cap on Plus; Pro tiers run on the weekly allowance only. Burn through it and Plus/Pro users can buy an instant weekly reset — it pulls your allowance forward rather than adding extra.
Why it matters
Picking the wrong tier is a tax on the uninformed. Pay for Pro when all you do is chat, and you've bought 5x–25x of an agent allowance you'll never touch. Stay on Free while running deep research daily, and you'll spend evenings staring at limit messages. Match the table row to what you actually do, not what you aspire to.
The context-window number matters too, for a subtler reason. "1.05M tokens" sounds infinite, and vendors love that. But knowing your usable share is smaller — and that a 200-page PDF plus "deep analysis" eats the window from both ends — is the difference between a workflow that works and one that mysteriously degrades at 2 a.m.
Impact on the USA and the world
With well over a billion people using ChatGPT weekly, these meters are among the most consequential pricing decisions in consumer software — they shape how students study and how small businesses automate. For US users: the Go tier may include ads, making ad-free Plus the real starting line for professional use; and every price is USD, billed monthly, with the Pro tiers ($100/$200/$500) aimed at researchers and heavy agent users.
What happens next
The direction of travel is clear: limits loosen as models get cheaper to run. OpenAI says it passes efficiency gains down to users — that's how everyday chat became unlimited — so expect the metered allowances to keep creeping upward while headline prices hold.
Our take: don't overbuy. Start on Free, and upgrade only when you can name the specific limit you're hitting — the reasoning cap, the upload ceiling, the weekly agent allowance. Most people never reach a wall. And when the API rate-limits post lands tomorrow, we'll cover the other side: what changes when you pay per token instead of per month.
API users: what's different over there
None of the above applies if you call the API with your own key. No plan, no message cap — you pay per token. Reference point: GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, per OpenAI's model docs. The constraint here is rate limits: requests-per-minute and tokens-per-minute ceilings that scale with your usage tier, plus the occasional 429 when you exceed them. Separate system — getting its own explainer tomorrow.
Frequently asked questions
Is ChatGPT really unlimited now?
Everyday text chat is — on every plan, Free included, subject to abuse guardrails, per OpenAI's pricing page. Reasoning models, image generation, uploads, voice, deep research, and the Work/Codex agent allowance all still have their own limits.
What exactly is a token?
A token is the chunk of text the model actually processes — roughly three-quarters of an English word. "ChatGPT token limits" as a phrase usually means the context window: how many tokens of conversation, documents, and instructions the model can consider at once.
Why does ChatGPT say my context is full when the window is 1.05M tokens?
Because the window is shared. OpenAI's own footnote explains that system instructions, tools, memories, and the model's reasoning all consume space from the same 1.05M-token pool — your input gets what's left. Long chats with memory and tools on fill up faster than the headline number suggests.
What's the difference between a usage limit and an API rate limit?
Usage limits cap a subscription over hours, weeks, or months (the 5-hour windows and weekly caps above). API rate limits cap requests per minute and tokens per minute for developers paying per token. Paying through an API key has no plan cap — but it has no free tier either.
Can I pay to reset my weekly limit?
On Plus and Pro, yes — OpenAI's help center describes an instant weekly Work/Codex reset you can buy from Usage settings. It restores your allowance immediately but pulls your normal weekly period forward; it doesn't stack extra usage on top. It's not available on Free, Go, Business, Enterprise, or Edu plans.
Related coverage
Confused about what's actually free in AI-land? Our AI-lane guide Cursor Is Free (But Not the Way Those "Free Cursor" Guides Claim) applies the same verify-the-tier approach to AI coding tools — free models, real limits, no marketing fog.
Sources
- OpenAI — Plans and Pricing | ChatGPT (chatgpt.com/pricing): plan features, unlimited-text-chat terms, shared-context-window footnote. Verified October 21, 2026.
- OpenAI — Models | OpenAI API (platform.openai.com/docs/models): GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna context windows (1.05M tokens), max output (128K), per-token API prices. Verified October 21, 2026.
- OpenAI Help Center — Paid weekly Work and Codex rate limit resets (help.openai.com/en/articles/20001507): instant-reset mechanics, plan eligibility. Verified October 21, 2026.
- OpenAI Help Center articles What is ChatGPT Plus? and About ChatGPT Pro tiers, plus OpenAI's DevDay 2026 recap (Sept 29, 2026): plan prices ($8/$20/$100/$200/$500), Pro usage multiples, GPT-6 Pro weekly caps — via corroborating October 2026 guides citing these official sources.