RN
Two numbers decide what one of these apps costs you, and the lists that rank them usually quote only the first.
The subscription is the number everyone prints. Underneath it there is almost always a second currency — tokens, credits, gems, coins, pick your noun — and that is what actually meters the features people sign up for. Images cost tokens. Voice messages cost tokens. Video responses cost a lot of tokens. The monthly fee buys you text and a token allowance, and the allowance is sized so that anyone using the app the way the marketing shows will run through it and buy more.
That structure is why a table of monthly prices tells you very little. Two apps at the same headline figure can differ by an order of magnitude in what a month of ordinary use costs, depending entirely on how the second currency is priced and how fast the interface spends it.
Why there is no ranking on this page
A ranked list of AI companion apps is a snapshot of a category that reprices, rebrands and reshuffles its tiers continuously. Publishing one implies every entry was checked, and checked recently.
This directory holds researched entries for three of the apps that turn up on those lists. It does not hold current, verified pricing for the rest, and inventing it would be worse than leaving it out — so what follows is the set of checks that produce your own ranking, plus what was actually established about the three.
The pre-signup disclosure test
The single most useful signal costs nothing to check: can you see the price before you make an account?
Some can. The researched entry for DreamGF found published monthly tiers sitting on the page, alongside tokens as the in-app currency — you can work out roughly what you are getting into before you commit anything.
Some cannot. The entry for Candy AI records monthly, quarterly and yearly plans and a separate token top-up, but the actual prices never appear before signup, and the site advertises a percentage discount without stating what the discount is from. The free tier's limits are not published either. That is a legitimate thing to be annoyed by, and it is a fair reason to move on.
Neither pattern makes an app good or bad at what it does. It tells you how the company treats the part of the transaction that costs you money, which is a reasonable proxy for a lot of else you cannot see.
Memory is the feature you cannot screenshot
Everything on a marketing page is a still image, and the thing people actually stay for is continuity: does it remember the name you gave it, the job you complained about, the thing that happened last Tuesday?
You cannot verify that from a gallery, from a demo conversation, or from anyone else's review — including this one. Continuity behaviour differs by tier on most services, changes when the underlying model is swapped, and degrades in long conversations in ways that only show up after a week.
Test it yourself, in the free tier if there is one, before the second billing cycle: mention something specific and unprompted, then come back to it days later in a fresh session. That single test separates the apps that hold a relationship from the ones that run a scene and reset.
Who wrote the character
There is a real split in this category that the ranked lists tend to flatten.
Some services ship a first-party roster: characters the company made, consistent in quality, limited in range. Others are publishing platforms where the characters come from users. The researched entry for GirlfriendGPT is a clear case of the second kind — its own navigation puts Top Creators third in the menu and ranks the people who make the characters, and its scenario tags run to twenty axes that mix relationship dynamics, fictional genres and body types on one row, because that is how the makers themselves tag.
The difference matters practically. A user-generated catalogue is enormous and wildly uneven. A first-party roster is small and predictable. Which one is better depends on whether you want to browse or to be handed something, and neither answer is wrong — but a list that ranks them against each other is comparing two different products.
The retention machinery
Look at what the app pays you to do. The DreamGF entry records a referral scheme that pays tokens for a signup and a daily claim bonus that rewards coming back, both restricted to paying users.
That is ordinary retention design, not a scandal. It is worth recognising for what it is, because it changes your usage in a direction that costs money: a daily bonus is an incentive to open the app on days you would not have, and an app you open more is an app you spend more in.
Before you pay
- Find the token price, not the subscription price. Work out what one image and one minute of voice cost, then multiply by how often the demo makes you want to press the button.
- Check whether the price is visible before signup. If it is not, that is information.
- Find the cancellation path before you subscribe, not after.
- Check how the charge appears on a statement if that matters to you. Candy AI's own FAQ has a specific answer about this, which tells you how often the question comes up.
- Test memory across sessions during the trial window.
- Note whether characters are first-party or user-made, and pick accordingly.
Where the checked entries live
The AI sex chat section carries the chat and companion services this directory has entries for, and AI porn sites covers the broader category including the image tools. If your interest is closer to roleplay than to a relationship, that guide covers the filtering question, which is a different problem from this one.