This catalogue is written for adults considering an AI companion subscription, and it is not intended for anyone under eighteen. Nothing here is a recommendation to subscribe; every decision belongs on the operator's own site, under their terms.
Candy AI arrives on this bench as a specimen, not a champion. Our console lays the product out across six measurable dimensions — how memory survives between sessions, what image work actually costs, whether voice exists on your tier, how far a single character can be shaped, where credits leak, and which kind of user the whole thing genuinely suits. Nothing here is a score handed down from above. Each station states what to observe, what stays unknown, and what to confirm on the sponsored partner page while the money is still in your pocket.
Laboratory-notebook framing: every block is a test area with observable signals, stated unknowns and a checklist, never a rating handed down from an authority figure.
Candy AI is one of the most searched companion products, and this page treats it as a research subject rather than a crowned winner. The honest takeaway is procedural: decide what you actually want from a companion, read the partner plan text yourself, run one deliberate session with notes, and only then judge whether a subscription matches your habits. A product can be genuinely well made and still be the wrong instrument for your bench — that mismatch usually reveals itself around week three, not on day one.
These six dimensions most often decide whether a companion subscription still feels worth paying for after the first month. Each is written as a station on the bench: what the dimension actually is, why it moves your experience, and what to confirm on the partner page before money changes hands. The order is deliberate — continuity first, because it is the quietest deal-breaker, then images, voice, character depth, the credit engine beneath the price line, and finally the fit question that no product page will answer on your behalf.
01
Continuity bench — what survives between sessions
Of all the dimensions on this bench, continuity is the one people notice last and resent most. Companion products split sharply on what carries over: a name, a stated preference, a running storyline, a mood, or nothing at all. Rather than trusting a headline about memory, treat retention as an open question you probe yourself. In practice that means referencing a small detail early in a conversation, returning later, and asking about it without a reminder. Whatever the result, it is data about your tier, your session length and the current model — not a permanent property of the product.
Continuity is the line between a companion that seems aware of you and one that greets you as a stranger each visit. It is the single most common reason people quietly stop renewing.
Confirm on the partner page: Read what the plan text says about context length, saved characters, and whether history survives a plan change, a logout or an account reset.
02
Image bay — output as a running cost
Image features rarely sit quietly inside a base subscription; they usually live behind a generation allowance, a credit pool or a premium tier. That makes them an economic dimension as much as an aesthetic one. Two questions matter here: how good the output ceiling actually is, and how fast a normal week of your usage would drain the allowance. A plan that sounds generous in text can become the expensive part of your month if most of your enjoyment comes from pictures rather than conversation. Watch for what happens mid-month when the pool empties.
A generous-sounding plan turns costly when images are the main reason you subscribe, because generation is often metered separately from ordinary chat.
Confirm on the partner page: Find the generation allowance on the plan page, check whether unused credits expire, and confirm what occurs when the limit is reached.
03
Voice channel — latency, availability, language cover
Voice is the station most likely to disappoint, because it splits into several distinct things that marketing tends to blur: spoken input, spoken output, real-time conversation, and language coverage. Availability also varies by tier, so a feature advertised for the product may not exist on the plan you are considering. If voice is the reason you are here, a text-only experience on your tier is a straightforward loss of value rather than a minor gap. Treat latency and language as things you can only observe in a live session, never infer from a feature list.
If voice motivates your subscription, discovering it is unavailable or metered separately on your tier is a direct reduction in what you paid for.
Confirm on the partner page: Verify whether voice means spoken input, spoken output or both, list the supported languages, and check whether it consumes separate credits.
04
Character range — breadth against depth
Catalogues compete on two different axes, and they reward different moods. Some win on sheer volume: hundreds of personas, easy discovery, constant novelty. Others win on how far a single character can be shaped — backstory, tone, boundaries, appearance, long-term arc. Breadth keeps the bench busy; depth makes one companion feel like a relationship instead of a slot machine. Neither is superior in the abstract, but choosing the wrong one for your temperament produces a very specific disappointment in week three: either boredom with the same voice, or frustration that nobody can be adjusted.
Breadth and depth are different pleasures, and a catalogue that excels at one can feel thin at the other, depending on whether you want variety or intimacy.
Confirm on the partner page: Decide which axis you actually want, then check how much a character can be edited after creation and whether edits draw on credits.
05
Credit engine — the economics beneath the price line
The advertised figure rarely describes the whole instrument. Downstream of it sit message limits, premium model access, image generations, voice minutes and optional top-ups, and those are what actually shape a month. Read the economy as a system rather than a single number: what a normal week of your own habits would consume, what happens when you run dry, whether boosters exist, and how refunds and cancellations are worded. Your true monthly cost is the plan price plus the extras your behaviour requires — and almost everyone underestimates their own behaviour while reading a plan page for the first time.
Real cost equals the plan plus the extras your habits demand, and habits are consistently the part people misjudge before they subscribe.
Confirm on the partner page: Estimate what one ordinary week would consume, ask whether top-ups exist, and read the cancellation and refund wording closely.
06
Fit matrix — matching instrument to user
Companion products reward different users differently. Story-driven roleplayers need continuity and editable characters; visual tinkerers need image allowance and consistent looks; casual chatters need low friction and predictable cost; prompt experimenters need range and responsiveness. A product can be genuinely good and still be wrong for you, and that mismatch usually surfaces in week three rather than at signup. Before comparing features, write your top two priorities on paper and hold them against the partner page. If a product wins on dimensions you do not care about, it has not won anything that matters to your bench.
A well-built product aimed at someone else's priorities will still feel like a bad purchase, and the disappointment tends to arrive after the first renewal.
Confirm on the partner page: Write your two highest priorities first, check only those against the partner page, and ignore the rest until both are answered.
Six partner paths
Bench index
The comparison below is a set of reading instructions rather than a leaderboard. Names in this category differ less in what they promise than in how their tiers meter continuity, images, voice and premium access. Read each product's own plan page with your two priorities in hand, and compare them against your habits rather than against each other.
This bench index gathers the dimensions that decide whether a companion subscription survives month two. Nothing is ranked. Each entry states what the dimension covers, why it shifts your experience, and what to confirm on the partner page before money moves. Read it as a checklist you carry to the plan page, not as a review you accept from us.
Step 1 — Write your use case before opening anything
Spend five minutes describing, in one sentence, how you would actually use a companion: short daily chats, long roleplay arcs, image-heavy creation, or voice conversation. That sentence becomes the filter you apply to every claim later. It also protects you from a subtle problem — marketing language setting your priorities before you have decided what they are. Keep the sentence somewhere visible, so that when a plan page argues a feature is essential, you can check whether it was essential to you ten minutes earlier.
Step 2 — Read the sponsored partner plan page
Open the partner site and read the plan structure itself rather than a summary of it: what each tier includes, what is metered, how billing recurs, and how cancellation is described. Note anything vague, because vagueness about limits is information too — it tells you which questions to ask before committing. Compare the tier you would realistically buy, not the flagship tier used in advertising. If the page will not state a limit clearly, assume the limit is tighter than the surrounding language suggests.
Step 3 — Run one deliberate first session
Start on a free or entry-level path if one exists. In that single session, test continuity by referencing a small detail early and returning to it later; watch how images or voice behave on your tier; and observe how the interface handles limits — whether it warns, blocks or silently meters. Take notes rather than impressions, because impressions smooth over exactly the details that decide a renewal. One structured session teaches you more about fit than an hour of browsing feature lists.
Step 4 — Apply the thirty-day rule before renewing
At the end of month one, compare what you actually used against what you paid for. If one or two dimensions carry the entire experience for you, subscribe for those and treat the rest as decoration. If volume is the only appeal, pausing is usually smarter than renewing on hope. This rule is unglamorous and it works: it converts a vague feeling of wanting to cancel into a concrete list of what you would miss, which is a much easier decision to make honestly.
Decision Matrix — Need, Signal, Tradeoff
Need
Look for
Trade-off
Long-form roleplay as the main use
How much context is retained, whether story state is saved, how deeply a character can be edited
Deeper continuity tends to sit on higher tiers, so the monthly cost can outrun what casual chatting would justify
Visual creation first
Image allowance size, editing options, consistency of a character's appearance over time
Allowances can drain quickly, and top-ups may become the largest recurring line
Voice conversation
Whether input and output are both spoken, which languages are covered, which tiers include it
Voice is often the most restricted feature and may be metered apart from text chat
Predictable, low spending
Clearly stated tier limits, no automatic top-up, short cancellation wording
Entry tiers give up premium models, faster replies or media features to hold the price down
Novelty and exploration
Catalogue size, variety, discovery and search tools, how often new characters appear
Large catalogues often feel shallower per character than tools built around one deep persona
Occasional, casual use
Monthly terms without lock-in and a straightforward pause path
Subscription pricing rarely rewards occasional users, so pausing usually beats any discount
Known Unknowns and Honest Limits
This catalogue is an editorial project, not a testing laboratory: we run no benchmarks, no comparative scoring and no user surveys, so nothing here should be read as a measured result.
Tier contents, feature availability and prices belong to the operator and can change at any time; the partner plan page always outranks anything written on this bench.
Memory behaviour shifts between sessions, models and tiers, so no retention claim — from us or from the operator — should be treated as permanent.
Content rules and moderation policies differ by product and evolve over time; what is permitted during your first session may not be permitted later.
This catalogue is written for adults, and every payment, subscription and cancellation happens on the operator's own site, under their terms rather than ours.
Frequently asked questions
I have never used an AI companion. Is Candy AI a reasonable first stop?
That depends less on the product than on your appetite for reading plan details. The conversation itself is generally described as approachable, but the value decision requires comparing what each tier meters and where the limits sit. If you prefer a guided start, work through the four protocol steps above before creating an account, and begin on the cheapest path that lets you observe behaviour. Starting cheap is not caution for its own sake; it is how you gather the evidence a subscription decision actually needs.
Can I judge the whole product from the free or entry experience?
Only partly. Entry access reveals a lot about interface feel, response style and how the product handles being interrupted, but it normally hides the layers paying users are actually buying — premium models, larger image allowances, voice, extended continuity. Use a free path to judge ergonomics and tone, not to draw conclusions about media quality or memory. The honest position is that entry access answers whether you enjoy talking to it and leaves the paid-tier question open until you read the plan page.
Why is there no score on this page?
Because a single number conceals the only thing that matters to you: which dimension you personally weight. Two people with identical habits can reach opposite conclusions about the same product, and averaging them produces a figure nobody can use. This page offers an inspection protocol instead of a ranking — stations, observable signals and questions to ask before paying. If you want a verdict, the honest one is that you are the only person positioned to produce it, and the protocol is simply how you get there.
How should I budget the first month?
Budget the plan price plus a deliberate buffer for whatever your own habits consume, because media generation, premium access and voice are commonly metered rather than included. Nobody can quote you a figure that would be honest, since it depends on your usage. Read the numbers on the partner page itself rather than trusting any figure quoted in an article, including this one. The practical goal is not a precise prediction but avoiding the unpleasant version of month two, where the total arrives higher than the headline you agreed to.
What is the quickest way to find out whether memory will disappoint me?
Reference a specific, slightly unusual detail early in a session — something the product has no reason to anticipate — then return later and ask about it without prompting. If continuity holds, it holds in a way a feature list cannot demonstrate; if it drops, you have learned the most important thing about the product for your purposes. Run that check before subscribing rather than after your first renewal, because memory is the dimension people most often assume and least often verify.
Are the outbound links here sponsored?
Yes, and the page says so wherever it matters. Outbound links to the partner are disclosed sponsorships, which is precisely why every claim above is written as a dimension to verify rather than an endorsement to accept. Sponsorship changes how this catalogue is funded; it does not change what the operator charges you, what their plan includes, or what their terms say. Treat the partner page as the authoritative document and this bench as preparation for reading it carefully.
