Home / Learn / Data & AI / Offline AI / Lesson 16.1
All lessons

Module 16: Cost: local AI against subscriptions · Lesson 16.1 · about 25 min

What Claude, ChatGPT and Google plans give you

The monthly AI plans from Anthropic, OpenAI and Google side by side in India, Canada, the USA and the UK, with what each tier gives, how the usage limits work, and the extras a local model does not have.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 16 · about 25 minutes · read first: 1.2 and 3.9. You’ll be able to compare the main AI plans in your country and know what you actually get for the money.

A US$20 AI plan is not unlimited: it is a metered allowance that resets every few hours, and its real price depends on your country, your tax and how many hours you actually use it.

1. Why this matters

Before you can say whether local AI saves money, you need the honest price of the other option. Plans look simple (“US$20 a month”), but they differ in limits, models, extras and price from country to country. Prices here were read on 19 November 2026. They change often, so check before you buy.

2. The plans, side by side

Pick your country:

Chart: cloud AI plans, per month
PlanPer monthWhat it gives
Claude ProUS$20Claude chat, Projects, Claude Code; usage resets every 5 hours with a weekly cap.
Claude Max 5xUS$100About 5 times Pro usage.
Claude Max 20xUS$200About 20 times Pro usage.
ChatGPT GoUS$8Lower-cost tier with fewer messages and smaller models.
ChatGPT PlusUS$20Full chat plus Codex; Codex messages are capped per 5-hour window (from about 5 to 3,000, by model).
ChatGPT ProUS$100 to US$500Tiers of US$100, US$200 and US$500; no 5-hour Codex limit.
Google AI PlusUS$4.99Gemini Pro model with daily caps, 400 GB storage, family sharing.
Google AI ProUS$19.99Gemini Pro with 1M-token context, daily caps, 5 TB storage.
Google AI UltraUS$100 to US$200Two tiers: about 5 times and 20 times Pro usage, Deep Think.
Checked 19 November 2026. India prices include GST. "Charged in US$" means no local price was published: your bank converts it and adds its fee, and sales tax or VAT is added at checkout. Sources: Claude Pro: claude.com/pricing; India: TechCrunch and Business Today, 13 to 14 July 2026; Claude Max 5x: claude.com/pricing; India: Business Today, 14 July 2026; Claude Max 20x: claude.com/pricing; ChatGPT Go: OpenAI Codex and ChatGPT pricing page; India: OpenAI, 2025 launch; ChatGPT Plus: learn.chatgpt.com/docs/pricing; UK: AI Tools Review, 2026; ChatGPT Pro: learn.chatgpt.com/docs/pricing; India: NewsBytes, 11 April 2026; Google AI Plus: The Tech Outlook, 9 June 2026 (price cut from US$7.99); Google AI Pro: Google blog, 19 May 2026; India: The Tech Outlook, 9 June 2026; Google AI Ultra: Google blog, 19 May 2026; India: The Tech Outlook, 9 June 2026. Prices and limits change often: check the official pricing page before you pay.

Three families, three tiers each:

  • Entry (ChatGPT Go, Google AI Plus): cheap, with smaller or capped models. Good for casual use.
  • Standard (Claude Pro, ChatGPT Plus, Google AI Pro): about US$20. The full everyday product, including coding tools (Claude Code, Codex) and longer limits.
  • Heavy (Claude Max, ChatGPT Pro, Google AI Ultra): US$100 to US$500 for 5 to 20 times the standard usage, early features and priority.

3. How the limits work

None of these plans is truly unlimited. Think of a phone plan with a data allowance that tops itself up every few hours. Each plan meters your use over a rolling window, often 5 hours, and most add a weekly cap.

What is really counted is computer work, and that follows tokens (Lesson 3.4). So a long document, a big coding task or a “thinking” model uses up the window much faster than a short chat. OpenAI publishes estimates: on Plus, between about 5 and 3,000 Codex messages per 5 hours, depending on the model. When you hit the limit you wait, switch to a smaller model, or upgrade.

9:00, the window opens. You have a full allowance for the next 5 hours. (Made-up numbers: call it 100 units.)

Short chats are cheap. Ten quick questions by 10:00 use about 2 units each: 20 units gone, 80 left.

A big task is expensive. At 10:30 you paste a long file and ask for a large code change with thinking on. In one hour it uses 60 units: thousands of tokens in, thousands out.

11:45, cut off. The last 20 units go on two more long requests. You are blocked until 14:00, when the 5 hours that started at 9:00 run out and the allowance refills.

The local version. A model on your own machine has no meter at all. Its only limit is how fast your hardware writes (Lesson 3.9).

4. Country notes

  • India: Anthropic, OpenAI and Google publish rupee prices, which include 18% GST. UPI works for some plans. With GST included, rupee prices land a little above the US list price.
  • Canada: most plans are charged in US$ and converted by your bank, with GST or HST added at checkout. Watch your card’s foreign-exchange fee.
  • USA: list prices are before state sales tax, which depends on where you live.
  • UK: some plans have pound prices (ChatGPT Plus at £20); others are charged in US$ with 20% VAT added.

The price you really pay is the list price with three multipliers on it: tax, the exchange rate and your bank’s fee.

5. What you get beyond the model

Cloud plans bundle extras that take work to copy at home: web search and deep research, file and project storage, voice mode, image and video generation, coding agents that run in the cloud, phone apps and syncing. Lesson 17.8 maps each one to its local version.

Plans also differ in data settings. Check whether your chats may be used to improve models, and switch it off if you prefer.

Maths and principle

Principle: compare plans per unit of what you actually use, not per month.

Equation: price per useful hour = monthly price ÷ hours of real use per month.

Analogy: a gym membership. Cheap if you go every day; expensive if you go twice.

Try it: a US$20 plan used 10 hours a month, and a US$100 plan used 80 hours? (US$2.00 an hour against US$1.25 an hour.)

price per hour=monthly pricehours used per month
monthly price
what leaves your bank account each month, after tax, exchange rate and card fees
hours used per month
the hours you really work with the AI, not the hours the app is open
rolling window
the few hours (often 5) over which your use is metered before the allowance refills
Physical meaning
A plan is a prepaid bundle of computer time in someone else's data centre. Used a lot, each hour is cheap; used a little, each hour is expensive. And in any one window you can only draw so much, however much you paid.

“I pay for the plan, so I can use it as much as I like.” You pay for an allowance, not for unlimited use. Each plan meters use in rolling windows and weekly caps, and long or “thinking” tasks drain them fastest. Heavy plans raise the ceiling; they do not remove it.

6. Worked example

Made-up usage, real list prices. Ananya in Bengaluru writes reports and some code. She pays ₹1,999 a month for ChatGPT Plus and keeps hitting Codex limits in the evening. Her options: Claude Pro at ₹2,000 for a second allowance, ChatGPT Pro at ₹10,699, or a local coding model on her 16 GB laptop for the routine work, so her Plus allowance lasts for the hard tasks. Lesson 16.6 shows why the last one is often the best value.

7. Practice

  1. Find your country’s price for the plan you use (or would use) and the date you checked it.
  2. What two things does a rolling usage window mean for your heaviest day?
  3. List three cloud extras you use that you would miss locally.
Answers
  1. Check the official pricing page at checkout; prices in this lesson may already have changed.
  2. You can be cut off mid-task, and big tasks use the window faster than chats.
  3. Commonly: web search, phone app, image or video generation.

New words: subscription tier, usage window, weekly cap, GST, HST and VAT, foreign-exchange fee.

  • Each family (Anthropic, OpenAI, Google) has entry, standard (about US$20) and heavy (US$100 to US$500) tiers.
  • No plan is unlimited: use is metered in rolling windows (often 5 hours) plus weekly caps, and long or thinking tasks use it fastest.
  • Your real price is the list price × tax × exchange rate × card fee; rupee and pound prices already include tax.
  • Compare plans by price per hour you actually use, not by the monthly sticker.
  • Plans bundle extras (search, voice, video, phone apps) that a local model lacks out of the box.

Check yourself

1What usually limits how much you can use a US$20-a-month plan?
2A plan has no price listed in your country's currency. What will you pay?
3Which of these does a cloud plan give that a local model does not, out of the box?

Sources

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.