How the pricing works
The free tier gives you a number of messages a day. Longer memory is a paid feature. There are credits, or gems, or energy, and a subscription that removes some limits but not all of them. The conversation sometimes stops at a well-chosen moment.
All of it makes immediate sense once you know the one fact the pricing is built around.
Every message costs money to answer
Not a fraction of a subscription — an actual per-message cost, paid by the operator, to whoever runs the model.
The cost depends on how much text goes in and how much comes out. Going in: the character description, the memory being carried, the recent conversation, your message. Coming out: the reply. Longer inputs cost more. Longer replies cost more.
That is a fundamentally different business from an app with a server bill. A companion app’s costs scale with how much people use it, and the heaviest users are the most expensive. A subscriber who sends several hundred messages a day can cost more to serve than they pay, which is why unlimited is rare and usually has an asterisk.
Once you hold that fact, nearly every design decision in the category is legible.
What the economics explains
Message caps on free tiers. A free user generates cost and no revenue. The cap is the point at which the operator stops paying for someone who has not paid them — set generously enough to let you form a habit and get attached.
Memory as a paid feature. Carrying more history means a longer input on every single message, so memory length is a recurring per-message cost rather than a one-off. Charging for it is arithmetic. It is also why memory shrinks when a subscription lapses.
Credits and currencies. Credits do two things a straight subscription cannot: they price heavy use without an unlimited promise, and they obscure the exchange rate. A pack of gems has no obvious relationship to a number of messages, which makes it much harder to notice how much a conversation cost.
Cheaper models on cheaper tiers. The most direct lever on cost is which model answers. Free and lower tiers routinely get a cheaper model, which is a substantial part of why the same character can feel different at different price points — and one of the reasons a character seems to change after a plan change.
Shorter replies. Output costs money. A product tuned toward brevity is cheaper to run, and that tuning lives in the character description where you cannot see it.
Slower responses at peak. Queuing paid users ahead of free ones is a cost decision presented as capacity.
The product
THE PRODUCT — pricing
· A message
→ a real per-message cost to the operator,
scaling with input and output length.
· Free-tier message caps
→ the point where they stop paying to
serve someone who hasn't paid.
· Memory sold as an upgrade
→ a recurring per-message cost, not a
one-off feature. Arithmetic, not
meanness.
· Credits, gems, energy
→ prices heavy use without promising
unlimited, and hides the exchange rate.
· "Unlimited"
→ check for fair-use or rate-limit
wording. Costs scale with use, so
something bounds it.
· Which model you get, how long memory is,
what a cap is, what it costs
→ THE OPERATOR DECIDES, and revises for
existing subscribers too.
· The actual cost per message
→ VARIES BY APP and is never published.The timing question, answered plainly
A paywall or a cap that arrives at an emotionally significant point in a conversation is the thing people most often ask about, so it deserves a straight answer.
The mechanism does not require anybody to be reading your conversation and choosing a cruel moment. It requires much less than that: engagement is measurable, and a company can see — in aggregate, statistically — that prompts shown after a certain amount of continuous use convert better than prompts shown at other times. Optimising for conversion finds those moments automatically, without anyone deciding what they mean.
The moments that convert best are the moments where you are most invested. That is not a coincidence and it is also not a plan; it is what optimising a number does. The useful consequence is not outrage, it is knowing that the timing of an offer carries no information about its value. An offer made at a compelling moment is the same offer at a boring one.
What to check before paying
What exactly the tier changes. Messages, memory length, model, response speed, filters — these are separate levers and a tier may move only one. The comparison table is often the only place a company states anything concrete about any of them.
Whether “unlimited” has qualifying language. Look for fair use, rate limits, or reasonable use in the terms. Costs scale with use, so something bounds it somewhere.
What credits convert to. If a purchase is denominated in a currency rather than messages, work out the per-message cost once. It is usually informative and it is deliberately not presented.
How to cancel, before subscribing. Whether cancellation is in the app or through the app store matters, because store-managed subscriptions are cancelled with the store and not with the app — a routine source of people paying for months longer than intended.
What happens to your history if you stop paying. Memory shrinking is common. Losing conversation access is less common and does happen, and it is worth knowing which one you are dealing with while you still have the account.
What this doesn’t tell you
It does not tell you what anything costs, at any app, because prices change constantly and nothing named here would stay true.
It does not tell you whether a subscription is worth it. That depends on what you get out of it, which is not a question this site is equipped to have a view on.
And it does not tell you that the design is malicious. Most of it is not: a business whose costs rise with usage will end up with caps and tiers whether it wants to or not. The reason to understand the economics is narrower — a limit you can explain is much easier to make a decision about than one that feels arbitrary.