AI Chat Costs: Subscriptions, Credits and Tokens

The monthly subscription price may be only part of what you pay for AI chat. A service might include a message allowance, sell credits separately, or charge for the amount of content a model processes. Voice and images can add another category.

To make a useful comparison, separate fixed fees from usage charges, then estimate the cost of your own habits. The worksheet below is designed for that decision. You do not need to become a developer to use it.

Find out what the purchase includes

Several categories can apply to the same product. A subscription that unlocks a feature does not necessarily cover all usage of that feature. Free installation does not establish the price of ongoing conversation.

Do not compare credit numbers across services. Ten credits in one app may buy a different amount of conversation from a hundred credits elsewhere. Compare the actual purchase cost of the same task, using the current offer shown to your account.

Why a short message can have a larger input cost

Tokens are units used to process content; they are not a fixed synonym for words or characters. Different kinds of input can be measured differently. Google's guide to understanding tokens explains the relationship between input, output and usage.

A request may contain more than your latest sentence: selected conversation history, character instructions and other context can be included too. The amount depends on the product. A ten-word message therefore does not establish the full input size.

Long chats do not inevitably become more expensive at a fixed rate. Products may shorten history, summarize it or use caching. Official model pricing can distinguish ordinary input, output and cache operations; a platform's selling price may differ from its supplier's price. See Anthropic's pricing documentation for an example of those categories.

Work through a hypothetical bill

These are invented teaching rates, not prices offered by Cakumi or any model provider. Assume a text-only service with:

• Input priced at $4 per million tokens.
• Output priced at $12 per million tokens.
• No cache discounts, extra feature charges or minimum charge.
• A request containing 8,000 input tokens and producing 500 output tokens.

Cost per request
= input tokens / 1,000,000 × input rate
+ output tokens / 1,000,000 × output rate

= 8,000 / 1,000,000 × $4
+ 500 / 1,000,000 × $12
= $0.038

At 40 identical requests a day for 30 days, the example totals $45.60. If input rises to 24,000 tokens per request while output stays the same, each request costs $0.102 and the example month totals $122.40.

The point is that context length can matter even when reply length stays constant. Neither figure predicts your bill. Real usage varies, and discounts, rounding, extra requests or other features can change the result.

Where detailed usage is available, add up the actual requests. Where only credits are shown, observe several ordinary sessions. A single short exchange is a poor basis for estimating a month of long conversations.

Keep a small personal usage record

Choose a trial budget you are comfortable spending. Record the tasks you actually care about, using this template:

Date:
Product, model and selected settings:
Task: everyday conversation / story / voice / images
Session length and number of replies:
Balance before and after:
Credits added during the session:
Actual charge or credits spent:
Regenerations or extra features used:
Did the experience meet my needs?

Account for credits added through purchases or rewards. Otherwise, a balance that barely changes can hide spending that occurred at the same time.

Also separate cash paid this month from credits consumed this month. You might be spending an older balance rather than making a new purchase.

Try a short conversation, a conversation with substantial history and any feature you regularly use. Those observations can suggest a personal budget range. They are not a claim about what everyone else will spend.

Reduce content you do not need

Change one thing at a time. Ask for shorter replies when a short exchange is what you want. Check for duplicated instructions. Disable optional features you are not using. Then compare both cost and conversation quality.

Removing all background information can create a different expense: repeated explanations and repairs. Before shortening a long thread, back up the original conversation. A new-chat handoff can help preserve selected context, although any summary loses detail.

The cheapest reply is not automatically the best value if you regenerate it repeatedly. A more expensive model is not automatically better for your preferred style either. Compare equivalent tasks, including the effort required to get a useful response.

Find your Cakumi model price and purchase breakdown

Rules reviewed on October 21, 2026. Cakumi measures AI usage in Resonance. Text model input and output prices are displayed in Resonance per million tokens. Keep the cash purchase, Resonance credited and chat consumption separate. The checks below help you understand your own cost; they do not promise a fixed monthly chat price.

1. Record the full selection in your Being's model settings. Note the model name, source and reasoning level, then the input and output prices shown in its expanded details. Switching the source of a model with the same name can change its price.
2. Check the asset in Wallet & Recharge. Separate general Resonance from model-specific Resonance. Credits allocated to one model are not a universal balance for every model. Record the asset actually used when comparing balances.
3. Read the final quote after selecting a product. Check the payment amount, base Resonance, fixed bonus, selected offer bonus and final credited total separately. Recheck the account, currency, product and offer before paying.
4. Separate recharge records from usage. A successful recharge confirms that credits arrived; it does not tell you the cost of a particular conversation. For a usage query, keep the time, model and source, feature, balance change and any available charge information. Contact the team if you need a charge checked.

Choosing English does not convert model prices into US dollars. Both cakumi.cn and cakumi.com show the web products available to the current account. App purchase options can differ by platform and installed version. A web price, an iOS purchase price and a provider's API price are not interchangeable: use the confirmation screen for the purchase route you are actually using.

Why one conversation can involve more than the main reply

The main text call in ordinary chat is priced using its complete input and output. Image understanding, memory retrieval, titles or summaries can also add usage, depending on which features actually run. Counting only your typed message and the visible reply can leave out context sent to the model and other calls.

Do not automatically apply a provider's cache discount to every Cakumi request. Ordinary chat uses complete-input pricing. For the first main call of a new Phone conversation round starting from idle, ordinary input and cache reads use the input rate, while cache writes and output use their respective rates; later calls in that round use the actual cache categories. If reported text usage cannot support the applicable calculation, local token estimates use the actual request and response text. An estimate is distinct from provider-reported usage. A charge should be checked against the usage, pricing method and user rates saved for that call.

Cakumi adds up the precise consumption within a settlement batch, then rounds the total up to a whole number of Resonance. This teaching example uses the actual rounding rule; the amounts are neither a customer bill nor live model prices:

The two calls are not rounded separately to charge 2 Resonance in this batch. If they belong to two separate batches, each batch is rounded separately. Speech recognition and delayed memory writes can settle separately, so one visible message does not establish that there is only one deduction. A batch totaling 1.1 Resonance is rounded up to 2 Resonance.

Compare recharge offers using the final total

CNY 1 = 100 Resonance describes custom-amount web recharges, not a universal conversion for every package, currency or payment channel. Standard packages use their own actual product prices, and bonuses affect the purchase cost per credit. Offers come from the quote for your account; do not add separate promotion percentages together yourself.

For the same payment, compare the confirmation screen's final credited amount rather than just the base number on a product card. Check that a voucher is valid and covers the selected product. This guide does not publish an unverified live ranking of the cheapest models, and the invented dollar rates earlier in the article must not be applied to Cakumi.

Make a small usage sample you can check later

Choose an ordinary task and hold the model, source, reasoning level and optional features constant for roughly ten exchanges. Measure a conversation with substantial history separately; a new, short chat does not represent every use case.

Date and start/end time:
Model / source / reasoning level:
Feature: ordinary chat / Phone / voice / images / other
Replies and regenerations:
General Resonance: starting __; ending __; added during session __
Relevant model-specific Resonance: starting __; ending __; added __
Other devices, conversations or delayed settlements active:
Observed use per asset = starting balance + credits added - ending balance

That balance difference only approximates the session's consumption when other activity has not mixed into it. Wait for delayed settlements to finish before taking the ending balance. If other conversations were active, do not attribute the entire difference to these ten exchanges. Several checkable samples with room for variation make a better personal budget than a promise that one recharge will last a fixed number of days.

For your initial decision, start with the kind of conversation you want: everyday companionship or character roleplay. That choice gives the budget a purpose.