Artificial Intelligence
How AI Subscription Pricing Actually Works: Tokens, Seats and Limits
A consumer plan sells a cap per rolling window. A team plan sells seats. The API sells tokens, and it charges for the whole conversation every time you send a message. Knowing which meter you are on is most of the cost.
By Maya Iyer · Updated · 10 min read
Short answer
Is any AI subscription actually unlimited?
AI products are priced on three different meters and the same vendor usually runs all three. Consumer subscriptions sell a usage allowance, typically a number of messages or an amount of compute per rolling window of a few hours, with higher tiers buying a larger allowance and access to stronger models rather than anything unlimited. Team and business plans sell seats, a fixed price per user per month with a pooled or per-user allowance and administrative controls. The API sells tokens, roughly three-quarters of a word each, priced separately for input and output, and every message in a conversation resends the whole conversation as input, so a long chat costs progressively more per turn. Reasoning or thinking tokens are billed as output. Rate limits on requests and tokens per minute are separate from spend and bind first for most small users. Cached-input and batch pricing are the two discounts that materially change API cost for repeated or non-urgent work.
Our practical verdict
The word subscription hides three different products. Understanding which one you are buying is most of understanding what it costs.
A consumer plan, the tier an individual pays for monthly, sells an allowance. Typically it is a number of messages or an amount of compute over a rolling window of a few hours, and when it runs out you wait for the window to refill or drop to a smaller model. A higher tier buys a larger allowance and access to the strongest models. What no consumer tier buys, whatever the plan page says at the top, is unlimited use of the best model; the fair-use terms further down are where the allowance is actually described.
A team or business plan sells seats: a fixed price per user per month, usually with a minimum count, and with it the administrative controls and data-handling terms that consumer tiers do not offer. For most small businesses the seat price is buying the terms rather than the capacity, and whether those terms matter is a question about what staff are pasting in rather than about how much.
Shortlist
Recommended options to compare
Use this as a starting list, then compare live India prices and warranty before buying.
Pick 1
Consumer plans meter a window, not a month
A paid consumer tier usually gives an allowance that refills over a rolling window of a few hours, not a monthly quota. Hit it and you wait or drop to a smaller model. Higher tiers buy a larger allowance and access to the strongest models. What no consumer tier buys is unlimited use of the best model, and the plan page usually says so in the fair-use terms rather than the headline.
Pick 2
Seats are the business meter
Team and enterprise plans charge per user per month, with a minimum seat count, admin controls, and the data-handling terms that consumer tiers lack. The seat price buys the controls and the terms as much as the usage. For a small Indian business the decision is rarely about capacity; it is whether the data terms matter, which the privacy guide covers.
Pick 3
The API meters tokens, in and out, at different prices
A token is roughly three-quarters of an English word and rather more than that for Indian languages, which tokenise less efficiently. Input tokens, what you send, are priced lower than output tokens, what the model writes. Reasoning or thinking tokens that a model produces before its answer are billed as output even though you may never see them, and on reasoning-heavy models they can dominate the bill.
Pick 4
Every turn resends the whole conversation
The API has no memory between calls. Each message you send includes the entire conversation so far as input, so the tenth turn of a chat costs roughly ten times the input of the first. This is why long-running agents and chatbots with long histories cost far more than the per-token price suggests, and why summarising or truncating history is the first optimisation anyone doing this at scale makes.
Pick 5
Rate limits bind before spend does
Requests per minute and tokens per minute are capped per account tier, separately from how much you are willing to spend. A new account on a low tier hits rate limits long before it hits a budget, and the way up is usage history and prepayment, not a larger card limit. For anyone building on the API from India, this is the constraint that surprises first.
Pick 6
The two discounts that matter
Cached input pricing charges a fraction of the normal rate for a prompt prefix the provider has recently seen, which is most of the cost in a long conversation or a repeated system prompt. Batch pricing takes roughly half off for work that can wait hours. Between them they can cut a repetitive workload's API bill by more than half, and neither applies to a consumer subscription.
Pick 7
Free tiers are the same meter, smaller
A free tier is a small allowance on a smaller model with the tightest rate limits and, usually, the weakest data terms. It is the right way to find out whether a tool fits a workflow. It is the wrong basis for judging cost, because the paid tier is metered the same way with the numbers moved, and the AI subscription cost guide is where the actual rupee comparison lives.
Which option should you choose?
Consumer plans meter a window, not a month
A paid consumer tier usually gives an allowance that refills over a rolling window of a few hours, not a monthly quota. Hit it and you wait or drop to a smaller model. Higher tiers buy a larger allowance and access to the strongest models. What no consumer tier buys is unlimited use of the best model, and the plan page usually says so in the fair-use terms rather than the headline.
Seats are the business meter
Team and enterprise plans charge per user per month, with a minimum seat count, admin controls, and the data-handling terms that consumer tiers lack. The seat price buys the controls and the terms as much as the usage. For a small Indian business the decision is rarely about capacity; it is whether the data terms matter, which the privacy guide covers.
The API meters tokens, in and out, at different prices
A token is roughly three-quarters of an English word and rather more than that for Indian languages, which tokenise less efficiently. Input tokens, what you send, are priced lower than output tokens, what the model writes. Reasoning or thinking tokens that a model produces before its answer are billed as output even though you may never see them, and on reasoning-heavy models they can dominate the bill.
Every turn resends the whole conversation
The API has no memory between calls. Each message you send includes the entire conversation so far as input, so the tenth turn of a chat costs roughly ten times the input of the first. This is why long-running agents and chatbots with long histories cost far more than the per-token price suggests, and why summarising or truncating history is the first optimisation anyone doing this at scale makes.
Rate limits bind before spend does
Requests per minute and tokens per minute are capped per account tier, separately from how much you are willing to spend. A new account on a low tier hits rate limits long before it hits a budget, and the way up is usage history and prepayment, not a larger card limit. For anyone building on the API from India, this is the constraint that surprises first.
The two discounts that matter
Cached input pricing charges a fraction of the normal rate for a prompt prefix the provider has recently seen, which is most of the cost in a long conversation or a repeated system prompt. Batch pricing takes roughly half off for work that can wait hours. Between them they can cut a repetitive workload's API bill by more than half, and neither applies to a consumer subscription.
Free tiers are the same meter, smaller
A free tier is a small allowance on a smaller model with the tightest rate limits and, usually, the weakest data terms. It is the right way to find out whether a tool fits a workflow. It is the wrong basis for judging cost, because the paid tier is metered the same way with the numbers moved, and the AI subscription cost guide is where the actual rupee comparison lives.
Choice IQ pick
Need the faster shortlist?
Start with our recommended options, then compare the tradeoffs that matter for your budget and workflow.
See top picksHow to decide
Pick the option around the job you need done. This is the fastest way to avoid paying for something that looks impressive but does not change your real workflow.
| Situation | Best starting point | Final check |
|---|---|---|
| Consumer plans meter a window, not a month | A paid consumer tier usually gives an allowance that refills over a rolling window of a few hours, not a monthly quota. Hit it and you wait or drop to a smaller model. Higher tiers buy a larger allowance and access to the strongest models. What no consumer tier buys is unlimited use of the best model, and the plan page usually says so in the fair-use terms rather than the headline. | Use this as a shortlist, then verify the final details before committing. |
| Seats are the business meter | Team and enterprise plans charge per user per month, with a minimum seat count, admin controls, and the data-handling terms that consumer tiers lack. The seat price buys the controls and the terms as much as the usage. For a small Indian business the decision is rarely about capacity; it is whether the data terms matter, which the privacy guide covers. | Use this as a shortlist, then verify the final details before committing. |
| The API meters tokens, in and out, at different prices | A token is roughly three-quarters of an English word and rather more than that for Indian languages, which tokenise less efficiently. Input tokens, what you send, are priced lower than output tokens, what the model writes. Reasoning or thinking tokens that a model produces before its answer are billed as output even though you may never see them, and on reasoning-heavy models they can dominate the bill. | Use this as a shortlist, then verify the final details before committing. |
| Every turn resends the whole conversation | The API has no memory between calls. Each message you send includes the entire conversation so far as input, so the tenth turn of a chat costs roughly ten times the input of the first. This is why long-running agents and chatbots with long histories cost far more than the per-token price suggests, and why summarising or truncating history is the first optimisation anyone doing this at scale makes. | Use this as a shortlist, then verify the final details before committing. |
Read the editorial notes
The API sells tokens. A token is roughly three-quarters of an English word and noticeably more than that for Hindi or Tamil, which tokenise less efficiently and so cost more per sentence. Input tokens, what you send, are priced lower than output tokens, what the model writes, because generation is the expensive step. Reasoning or thinking tokens, which some models produce before their answer, are billed as output whether or not you see them, and on reasoning-heavy work they can be most of the bill.
The mechanic that catches nearly everyone building on the API is that the model has no memory between calls. Every message you send includes the entire conversation so far as input. The tenth turn of a chat carries ten turns of history, so the cost per turn climbs through a conversation, and a chatbot or agent with a long running history costs a multiple of what the per-token price implies. Summarising or trimming history is the first optimisation anyone doing this at scale makes, and cached-input pricing, which charges a fraction of the rate for a prompt prefix the provider has recently seen, is the pricing feature built for exactly this. Batch pricing, roughly half off for work that can wait hours, is the other discount that materially changes a repetitive workload's bill.
Before spend binds, rate limits do. Requests per minute and tokens per minute are capped by account tier, independently of budget, and a new account on a low tier hits those caps long before it hits a number on a card. The way up is usage history and prepayment. For anyone building from India this is the constraint that arrives first and is least expected.
A free tier is the same meter with smaller numbers: a modest allowance on a smaller model, the tightest limits, and usually the weakest data terms. It is the right way to find out whether a tool fits a job and the wrong basis for judging what the job will cost, because the paid tier is metered identically with the figures moved. The actual rupee comparison across plans lives in the AI subscription cost guide, and what an Indian card statement adds on top is its own subject in the paying for AI from India guide.
Decision shortcut
Still comparing options?
Use the table above to shortlist your best fit, then check related picks, tools, and buying guides before you make the final call.
FAQ
Is any AI subscription actually unlimited?
No. Consumer tiers sell a usage allowance that refills over a rolling window of hours, with fair-use terms; higher tiers buy a bigger allowance and better models. Business seats pool or cap usage per user. The API meters every token.
What is a token and why are input and output priced differently?
A token is roughly three-quarters of an English word, and more for Indian languages. Output tokens cost more because generating text is the expensive step; input is what the model reads. Reasoning tokens are billed as output.
Why does a long chat on the API cost more per message?
Because every message resends the whole conversation as input. The tenth turn carries ten turns of history, so cost per turn rises through a conversation unless history is summarised or trimmed.
Why am I rate-limited when I have budget left?
Requests and tokens per minute are capped by account tier, independent of spend. New accounts sit on low tiers and move up with usage history and prepayment, not with a higher card limit.
Which API discounts are worth knowing about?
Cached input, which charges a fraction of the rate for a recently seen prompt prefix, and batch processing, which discounts work that can wait hours. Together they can more than halve a repetitive workload's bill.
Features, model access, limits, privacy controls, and India billing can change. Choice IQ evaluates practical workflow fit and recommends checking the provider's current official terms before paying.
AI Tools Editor
Maya reviews AI products, productivity systems, and automation workflows with a focus on practical adoption.
Keep reading
Related artificial intelligence picks
Comparison
AI API Pricing 2026: What Every Model Actually Costs
Every current model's token pricing in one table, including the promotional rates that expire and the long-context rates that catch people out.
Comparison
AI Subscription Cost in India: ChatGPT Plus vs Claude Pro vs Gemini
What the three main AI subscriptions actually cost an Indian buyer per month, including the billing detail that adds to one of them.
Artificial Intelligence
What 'Unlimited' Means on an AI Plan
No AI plan is unlimited. The word means a cap you are unlikely to hit under normal use, a fair-use clause that defines normal, and a smaller model waiting when you exceed it. Here is where the actual limit lives.
Artificial Intelligence
AI and Indian Languages: What Genuinely Works Today
Machine translation between Indian languages and English is genuinely good. Generating a Hindi email is fine. Transcribing a Bengali call is usable. Writing long-form Tamil with any literary judgement, reading a handwritten Marathi form, or handling a Bhojpuri speaker is not there yet, and Indic text costs two to four times the tokens.
Tool
AI Use Case Finder
Identify a sensible first AI workflow based on your role, repeated task, and risk level.
The best choice is rarely the product with the longest feature list. It is the one you will still trust and use six months from now.
How Choice IQ evaluated this guide
Choice IQ explains AI pricing by the three meters vendors use - windowed consumer allowances, business seats, and API tokens - and the mechanics that drive real cost: input versus output pricing, reasoning tokens, conversation history resent per turn, rate limits independent of spend, and the cached-input and batch discounts.
Weekly intelligence
Get smarter AI tool picks weekly.
One useful email with practical comparisons, refresh alerts, and decision frameworks for people who do not want another noisy newsletter.