ChatGPT Plans Compared (2026): Free, Go, Plus, Pro and Business

On 3 October 2026, ChatGPT sells nine ways to pay, and the cheapest paid one has no dollar price on OpenAI's own help page. The free tier costs nothing. The $20 plan covers most people who pay. Above it sit three monthly plans for heavy users, and below the seat prices for teams, a quote for large firms.

Choosing got harder on 29 September 2026, when OpenAI cut what the $200 plan includes for new buyers and added a $500 plan. If you read a pricing post from last month, parts of it are already wrong.

This guide is for one person deciding what to pay for, or for a team lead choosing seats for two to twenty people. Every figure comes from the pricing page or an OpenAI help page, all opened on 3 October 2026, and the text says where those pages leave a gap. Short version: stay free until a limit stops you, then pay for the $20 plan. Treat the three top plans as tools for people who run long agent jobs every day.

How to Choose: Four Questions Before You Pick

Answer these in order. Stop at the first one that settles it.

1. Did a usage limit stop real work in the last two weeks? If not, stay on the free tier.
2. Do you paste long documents or whole code files? Check the context table below before paying for more.
3. Do you run Codex or agent jobs for hours at a time? That is the only case for the $100 plans and up.
4. Will a team share the account? Then buy workspace seats, not personal plans, because the data terms differ.

The Shortlist: Which Plan Fits Whom

Read this before the detail.

• A few questions a day: the free tier, which allows unlimited text chats.
• Daily use with more uploads, images and voice: Go, where it is sold at a price you accept.
• Regular work, Codex sessions and deep research: the $20 plan.
• Hours of agent work every day: one of the three top plans.
• A team that needs a shared workspace and no training on its data: Business, from two seats.
• A firm that needs audit logs, key management and data residency: Enterprise, by quote.

What Each Plan Costs

Prices are per month in US dollars unless stated. The $20 figure comes from the help page for that plan, the three top plans from the tiers page, and seats from the business pricing page. OpenAI says the individual plans have no annual option.

Two things in that table need a warning.

First, the price you see depends on where you are. The page I opened from Indonesia listed Go at IDR 75,000 and Plus at IDR 349,000. The help page for Go says purchases are in US dollars, with local billing in a limited set of countries. It then points you to the pricing page for the figure and prints none itself.

HokAI's own ChatGPT listing last recorded Go at $8 a month in the United States, from a July check. Treat that as a lead, not a quote.

Second, Go is the plan most likely to change. OpenAI says it may adjust pricing as it learns from early use, and the pricing page notes that Go may carry ads. If ads in a paid plan bother you, skip it.

The Free Tier: What You Get Without Paying

The free tier gives unlimited text chats on GPT-5.6 Luna, subject to abuse guardrails. Everything else is capped. Uploads, image creation, voice, deep research, memory and Codex access are all labelled "limited" on the pricing page. It also gets limited access to ChatGPT Work, on the desktop app only.

That is more than it sounds. For drafting, summaries and the odd question, a person may never hit a wall. The wall arrives on the first long document, the first batch of images or the first hour of voice.

The context window is the other limit. OpenAI lists 27K tokens for everyday chat here, about 12 pages of pasted text. A long contract will not fit in one go.

Go: The Cheap Middle That Is Easy to Misjudge

Go sits between free and $20 and is built for longer conversations. OpenAI lists more messages with tools, more uploads, more image creation, more voice chats and longer memory than the free tier. Projects, tasks and custom GPTs are included, subject to storage limits.

Two limits catch buyers out. Go includes a Think mode, which runs on Luna, but it does not include GPT-5.6 Sol. It also has no legacy models. If a favourite workflow depends on an older model, this plan will not bring it back.

The context figures are the surprise. Go and the $20 plan list the same 54K everyday window and the same 256K reasoning window. So the extra money buys models and allowance, not a bigger window. Judge the two plans on what you do, not on how much text you can paste.

Plus at $20: Where Most Paying Users End Up

OpenAI describes this plan as broader model and tool access than the free tier, with priority access in busy periods and faster replies. The pricing page adds advanced reasoning with the newer model family, expanded deep research and expanded access to Codex and Work on desktop, web and mobile.

Here is what the move up from Go buys in practice:

• Reasoning models in the newer family, such as GPT-6 Sol and GPT-6.1 Sol, in the tools where OpenAI has switched them on.
• Expanded rather than extended limits for deep research, memory and uploads.
• More complex image creation, which is where Images 2.5 comes in. HokAI's record says OpenAI released it on 8 September 2026 as two API snapshots and routes each prompt to the better fit. It replaced GPT Image 2, released on 21 April 2026.
• Early access to new features.

OpenAI also warns that availability moves during rollouts. Check the model picker in your own account rather than trusting any list, this one included.

The Top Three Plans, and a Change You Should Know About

The top end is now three monthly plans. They differ in price and in what they include.

Only the $500 plan has Ultrafast, the fastest speed mode for GPT-6 Astra. That speed costs allowance. The help page says Fast mode spends your included usage at 2.5 times the standard rate, and Astra Ultrafast at 8 times. Those are consumption rates, not promises of speed.

The change affects timing. New $200 subscriptions are on sale again, but with a lower allowance than before. Anyone who held that plan from 22 September to 10 a.m. Pacific on 29 September 2026 keeps the old allowance until 29 October. After that the price stays put and the allowance drops.

Secondary coverage puts numbers on it. One news summary says the top plan carries 25 times the allowance of the entry paid plan, and that the middle plan fell from 20 times to 10 times in Codex and Work. OpenAI's help page does not print those multiples, so treat them as reported, not confirmed.

The plain consequence: a review that praised the $200 plan's allowance may describe something you can no longer buy. The price is unchanged, so a new buyer gets less for the same money.

Credits are the release valve. You can buy credits to carry on after the included allowance runs out in Codex, Work and the Word, Excel and PowerPoint add-ins. They do not cover every model, and at launch they do not add Ultrafast to the two cheaper plans.

The pricing page also lists "Dot", an always-on agent, unlimited and faster image creation, and the longest Codex and Work sessions among the top-tier perks. For what that agent does, see what OpenAI's Dots do and the agent page.

Business and Enterprise: Seats, Not Plans

Business is sold per seat, from two seats up. Each seat type has a monthly and an annual rate, shown in the first table. OpenAI says Premium carries 5 times the usage of Standard with no 5-hour limit. Teams can mix both seat types in one workspace.

Annual billing is the one place on this page where committing saves real money. The arithmetic for five seats:

Those figures are my arithmetic on list prices, not a vendor quote. A team of three Standard and two Premium seats comes to $3,120 a year on annual billing.

Business includes single sign-on, central billing, usage analytics and no training on your data by default. Enterprise adds the heavier controls on the pricing page: SCIM, key management, role-based access, an analytics dashboard and data residency in ten regions, among them the US, the EU, Singapore and India. It is quote-only.

Two discounts sit outside the table. OpenAI offers verified US K-12 educators a free plan through June 2027, and says nonprofits can get up to 75% off Business or Enterprise.

How Much Text Each Plan Can Hold

Context window is the figure plan pages usually hide, and OpenAI does print it. This is the shared window for everyday chat and for reasoning, as listed on 3 October 2026.

The input limits are smaller than the windows. OpenAI explains that system instructions, memory and the model's own reasoning use part of the space, so the room left for your text is an approximation that shifts with the features in use. A 300-page report on the $20 plan is close to the ceiling, not comfortably inside it.

For most users the jump that matters is the top tier, where the reasoning window grows to 400K. If you paste whole code repositories or long legal bundles into one chat, that is a real reason to pay more. For everything else it is not.

Which Models Come With Which Plan

OpenAI's model line-up changes more often than its prices, so read this section as a map, not a promise.

• The free tier runs on Luna for text chats.
• Go adds a Think mode on the same model, but not Sol.
• The $20 plan lists advanced reasoning with the newer family. Check the picker to see which models your account has.
• The top tier adds Pro reasoning powered by Astra.
• GPT-5.6 Terra appears in the comparison, with limited access in Work and Codex on desktop shown for two of the plans.

HokAI's model record describes Terra as the balanced tier of its generation, with a 200K context window and no ultra mode. It matters here because it is the lower-cost model named for Work and Codex jobs.

If you also use the API, none of this applies. API usage is billed separately and is not included in any consumer plan, according to the help pages for Go and the $20 plan. For API prices, see the model leaderboard.

The Turn: Maybe You Should Pay for Nothing

Here is the best argument against paying at all. Unlimited text chats on a capable model cost nothing. A person who writes emails and asks questions all day may never touch an image or a long upload. For them, $20 a month buys priority and polish.

That argument is right more often than sellers admit. The test is simple. If no limit has blocked you in two weeks, you are not short of plan.

It fails in three cases. The first is the small window, which blocks long documents. The second is Codex or deep research, where "limited" runs out in an afternoon. The third is image work, where one project can use up the free allowance.

If any of those describes your week, move to the $20 plan. If none does, wait.

Who Should Skip the Paid Plans

Paying OpenAI is not automatic, even for a heavy user. Three groups do better elsewhere.

Developers who spend the day in an editor should compare a coding tool first. GitHub Copilot is built around the editor, and the coding assistant roundup shows how the main options differ.

People who mostly need cited research may prefer a search-first assistant, and teams that live in Google Workspace may find Gemini fits better. Our Gemini explainer covers that. Microsoft shops can look at Microsoft Copilot.

For a straight head-to-head, the guides on ChatGPT versus Grok and ChatGPT versus Kimi compare the assistants, not the plans. The wider assistant choice guide covers five rivals at once, and the chatbot category lists the rest.

What Happens to Your Data on Each Plan

The pricing page marks every individual plan with the same line: content may be used to train the models, with an opt-out available. The business pricing page says the opposite for work accounts: no training on your data by default.

That split decides the question for anyone handling client material. Pasting a contract into a personal account is a different act from pasting it into a workspace, even when the model is identical. If your employer pays for seats, use them for work. If you use a personal plan for work, turn the training opt-out on first and check that your employer allows it.

Enterprise adds data residency in ten regions and the compliance certificates on the pricing page. For regulated work, ask for those documents rather than relying on a plan name.

Switching, Cancelling and Refunds

Changing plan is easy. Open the settings page and look for your plan. A downgrade does not refund the days left on the old plan: if you move from the $20 tier to Go, the current plan runs to the end of the billing cycle, then the Go rate applies.

Cancelling works on the platform where you bought. If you subscribed through Apple or Google Play, cancel there, since deleting the app does not end the subscription. Scheduling a cancellation keeps access to the end of the period, and you can undo it before then.

There is no annual discount on individual plans, so there is no lock-in either. That is a quiet advantage. You can buy a top plan for the one month you need it and drop back for the rest of the year.

For a team, the lock-in runs the other way. Annual billing is cheaper but commits you for the year, so start with monthly seats for a pilot, then switch once usage is steady.

What Would Change This Verdict

This guide is a snapshot of one day, and plan pages are among the least stable pages on the web. Three signals would change the advice.

If a dollar price for Go appears on the help page, the cheap-plan advice gets simpler. If the $200 allowance moves again after the 29 October cutoff, the middle plan may stop being worth its price. And if the Chat tab starts running the newest models for $20 subscribers, the gap between Go and that plan will widen.

To see how the models behind each plan compare on price and scores, use the live model leaderboard. If you are weighing the plans against another assistant, HokAI's Smart Match starts from your workload and narrows the field. For the wider picture of what people do with the assistant, see ChatGPT use cases and the guide to choosing a large language model.

Frequently asked questions

How much does ChatGPT cost per month in 2026?

Free costs nothing and Plus is $20 a month. Pro comes in three plans at $100, $200 and $500. Business seats are $25 monthly or $20 billed annually for Standard, and $125 or $100 for Premium. Go is priced by country, and OpenAI's help page no longer prints a dollar figure for it.

Is ChatGPT Plus worth it over the free plan?

It is worth it once a limit blocks real work. Free allows unlimited text chats but caps uploads, images, voice, deep research and Codex. If you have not hit a cap in two weeks, wait. If long documents or coding sessions keep stopping you, Plus is the usual next step.

What is the difference between Pro 100, Pro 200 and Pro 500?

Price and included usage. Pro 200 includes more than Pro 100, and Pro 500 includes the most. Pro 500 is the only plan with Ultrafast speed for GPT-6 Astra. New Pro 200 buyers get a lower allowance than before, while earlier subscribers keep theirs until 29 October 2026.

Does ChatGPT train on my data on paid plans?

On the individual plans, Free through Pro, OpenAI's pricing page says content may be used for training, with an opt-out available. Business workspaces default to no training on business data. Check the data controls in your account before pasting client material into a personal plan.

Can I pay for ChatGPT annually?

Only for Business and Enterprise. OpenAI says Go, Plus and Pro have no annual option and cannot be paid for months in advance. Business seats cost $5 to $25 less per month on annual billing, which saves a five-seat Standard team $300 a year.

Covered in this guide

• ChatGPT: ChatGPT is OpenAI's AI assistant, with 1.2 billion weekly users by OpenAI's count, the GPT-6 model family and plans that run from a free tier to a premium Pro tier.
• Images 2.5: OpenAI's ChatGPT Images 2.5, released September 8, 2026, adds Sketch reference drawing and comment-based edits across two API snapshots, Flare and Sunburst.
• ChatGPT Work: OpenAI's ChatGPT Work agent plans and finishes multi-step projects across connected apps and files. It launched on 9 July 2026 and now reaches every ChatGPT plan, with limits.
• Gemini: Google's multimodal AI assistant and model family, sold as a free app, tiered subscriptions and a pay-per-token API
• GitHub Copilot: GitHub's AI coding assistant with agent mode and MCP support across 6 IDEs, plus a free plan with no credit card required.
• GPT Image 2: OpenAI's April 2026 image flagship integrates O-series reasoning for 99% text accuracy, natural language inpainting, and 2K resolution via a token-based API.
• GPT-5.6 Luna: GPT-5.6 Luna by OpenAI (July 2026): budget-tier pricing, fastest in family, high-volume workloads. 200K context, vision plus text, capped reasoning.
• GPT-5.6 Terra: GPT-5.6 Terra is OpenAI's balanced-tier model, released July 2026 with strong reasoning, a 200K context window, and no ultra mode.
• GPT-6 Astra: OpenAI's flagship model, launched September 2026 as the first ever rated at the Preparedness Framework's Critical cybersecurity level.
• GPT-6 Sol: OpenAI's mid-tier GPT-6 model (Sep 2026), 1.05M-token context, 68.8% on DeepSWE v1.1 coding, half GPT-5.6 Sol's price.
• GPT-6.1 Sol: GPT-6.1 Sol is OpenAI's September 2026 mid-tier reasoning model for coding, computer use and document work, built to sit just below GPT-6 Astra.
• Microsoft Copilot: Your everyday AI companion powered by GPT-5.6 (preferred model July 2026), GPT-4o, and GPT-5 for productivity, creativity, and intelligent assistance
• OpenAI: OpenAI builds ChatGPT, the GPT-6 family (Astra, Sol, Luna), Codex and the OpenAI API. It reports 1.2B weekly ChatGPT users and an $852B valuation after its March 2026 round.
• Codex: Clones a GitHub repo into a sandbox, writes and tests code, then opens a pull request; OpenAI reported 20 million active users in August 2026.
• agent: OpenAI's always-on agent in ChatGPT runs on GPT-6 Astra with its own cloud computer, and asks approval before consequential actions.

Sources