What AI roleplay costs: a subscription against your own API key

Last updated September 19, 2026 · Any AI character chat app · By the developer of LonelyTavern

Cancel a monthly subscription, pay a model provider directly instead, and do you spend more or less? Here is the arithmetic, at real prices, so you can run it on your own numbers.

The short answer

• A subscription is a flat fee. Your own key is a meter. That is the whole difference, and it is why paying per message feels risky.
• The meter has a hard ceiling if you want one. Most providers sell prepaid credit. Put in $10 and $10 is the most you can spend. When it runs out the app stops, and nothing bills you again.
• The model you pick decides everything. On a cheap model, one month of subscription money buys thousands of messages. On a top-tier model it buys a few hundred.
• You can also spend nothing. Free models exist, with daily limits and a privacy cost. More on that below.

Why a message costs anything at all

The model remembers nothing between messages, so the app resends the conversation every turn. That is also why characters forget things. It means you pay for the whole history on every reply, not only for the words you typed.

So two numbers set the price:

• Input. Everything sent to the model: the character card, your settings, and as much of the conversation as still fits.
• Output. The reply that comes back. A token of output costs several times more than a token of input, but there is far less of it.

A token is about three quarters of a word. The examples below use an 8,000-token conversation and a 300-token reply, which is an ordinary roleplay turn.

What one message costs

• A budget model at $0.30 per million input tokens and $1.20 per million output: about $0.003 a message. One dollar is roughly 350 messages. This is the tier most roleplayers actually use.
• A top-tier model at $2 input and $10 output: about $0.02 a message. One dollar is roughly 50 messages.
• A free model: nothing, up to a daily request limit.

Seven to one, for the same conversation. That gap, not the billing method, is what decides whether paying by the token beats a subscription.

Two things push the real bill below that arithmetic:

• Caching. Most providers charge a fraction of the input price for the part of the conversation they have already seen, which is most of it. On the cheapest models that fraction is a fiftieth, so a run of quick replies costs far less than the figures above.
• A shorter context. Trimming what the app resends saves money on every turn. It is the one setting that really changes your bill.

Find the number you pay now

Character chat apps cluster around four prices. Find the one on your receipt, and the figures show what the same money buys as prepaid credit at the two ends of the model range.

• About $5 a month. The stripped entry tier, and the one most likely to still meter your messages. The same money as prepaid credit: roughly 1,800 messages on a budget model, or 260 on a top-tier one.
• About $10 a month. The mass-market tier. Ads off, queue priority, usually no message cap. The same money: roughly 3,600 budget messages, or 520 top-tier ones.
• About $14 a month. The full-feature tier, and the most common price in this market. The same money: roughly 5,000 budget messages, or 730 top-tier ones.
• About $25 a month. The power tier, sold on longer memory and better models. The same money: roughly 9,000 budget messages, or 1,300 top-tier ones.

For scale, 1,800 messages a month works out to 60 a day. That is a long evening of roleplay, every day.

When the subscription is the better deal

• You want a top-tier model and you chat for hours every day. Past roughly 700 messages a month on one of those, a $14 plan is cheaper than the meter.
• You use the extras. Voice, images, and video are bundled into some plans and billed separately everywhere else.
• You do not want another account. A key means signing up with a model provider and topping up before the first message.
• You want one bill you never think about. A meter asks you to glance at it occasionally, even when it is small.

When your own key is the better deal

• You are on a capped plan. Message credits and daily allowances are the subscription's version of a meter, and you cannot see it running.
• You send fewer than about 2,000 messages a month. On a budget model that costs under $6, not fourteen.
• You want a model the app does not offer. With your own key the model list is the provider's, not the app's, and you can change it mid-conversation.
• You do not want the rules changed on you. A subscription app can retune its model, tighten its filters, or reprice its tiers. Your key keeps working.
• You want a hard ceiling. Prepaid credit is the only plan that literally cannot charge you again next month.

Keeping the meter low

• Start with $5 or $10. Do not connect a card for automatic top-ups until you know what a week actually costs you.
• Watch the first week. Every provider has a usage page. After seven days you will know your real monthly number, and it is usually lower than you feared.
• Cap the context. The single biggest lever. Resending 4,000 tokens instead of 12,000 cuts the input cost by two thirds on every turn.
• Use the cheap model for ordinary scenes. Save the expensive one for moments you care about. Swapping models mid-chat costs nothing.
• Set a spending limit on the key where the provider offers one, and use a separate key for each app.
• Do not stockpile credit. Several providers expire unused credit a year after you buy it. Top up what you will spend in a few months.

The free route, and its price

You can run a character-chat app without paying anything. Two ways, both with a catch:

• Free model variants on a router. Real models at $0. On OpenRouter that means 20 requests a minute, and 50 a day until you have bought $10 of credit in total, after which it is 1,000 a day. Enough for casual chatting, tight for a long evening.
• A provider's free tier. Generous daily limits, and the provider may read and train on what you send. Paid usage at the same company usually is not treated that way.

Free is the right way to test an app before you top anything up. Read the provider's terms before you put a story you care about through it.

Related guides: Which API key do you need for AI roleplay · Why your AI character forgets things · How to write a character card that works · all guides.

Sources

• Subscription prices: each app's own US App Store listing and published pricing page, read on September 19, 2026. Prices are grouped into tiers here rather than attributed, because the tier is what the arithmetic needs.
• DeepSeek: models and pricing: the $0.30 and $1.20 budget-tier figures, and the cache-hit price
• Anthropic: pricing and OpenAI: API pricing: the $2 and $10 top-tier figures
• DeepSeek: multi-round chat: "The user must concatenate all previous conversation history and pass it to the chat API with each request", which is why you pay for the history every turn
• OpenAI: key concepts: one token is about four characters, or three quarters of a word
• OpenRouter: API rate limits: the free-model request caps

The apps described by price are independent products of their respective companies. LonelyTavern is not affiliated with or endorsed by any of them, and earns nothing from what you spend with a model provider. Prices change; check the current ones before you decide.