The category stopped growing outward this year and started splitting in two.
A year ago the annual prices in this category were scattered. Now most of them sit between $5.83 and $9.99 a month on a twelve-month plan: SweetDream AI at $5.83, Replika at $5.83, Candy AI and Secrets AI at $5.99, SoulGen at $7.58, Nomi AI and Janitor AI Pro at $8.33, OurDream AI at $9.99.
Convergence like that usually means a category has found its price, and it means something specific for buyers: the interesting differences are no longer in what a plan costs but in what it meters. A $5.99 plan with token costs on top and a $9.99 plan with none are not comparable products, and the headline number no longer distinguishes them.
The outliers are informative. Kindroid at $9.99 to $13.99 declines to discount annually at all. Seduced AI starts at $25 with no free tier. PornPen sits at a flat $15 with no metering. In each case the pricing structure is telling you what the product thinks it is.
The clearest finding across this year's testing is that chat-first and image-first platforms have stopped converging and started specialising.
At the chat end, Nomi AI scores 9.1 on conversation and 6.8 on images. Kindroid scores 9.0 and 7.4. At the image end, SoulGen scores 8.5 on images and 2.0 on chat; PornPen 7.8 and 1.0; Seduced AI 8.0 and 1.5. These are not versions of the same product with different emphases. They are different products.
The hybrid platforms in the middle β OurDream AI, SweetDream AI, Candy AI β are the ones doing both, and they are genuinely good at both without leading either. Candy AI's 8.7 on images is the highest score on this site, and its 8.0 on chat sits a full point below Nomi. That is the trade the middle of the category makes, and it is a reasonable one if you want a single account.
What has changed is that the trade is now visible. A year ago every platform claimed to do everything; this year the pricing pages have started admitting what they are.
Memory has the widest spread of any axis we score: 9.3 for Nomi AI at the top, 1.0 for PornPen at the bottom, with most of the companion platforms clustered between 6.0 and 8.0.
More importantly, it is the axis that predicts whether someone is still using a platform in month two. Conversation quality in week one is a function of the underlying model, and the underlying models are increasingly similar. Conversation quality in week four is a function of whether the platform kept anything, and that is entirely an architecture question.
Nomi AI and Kindroid are the only two here that passed our thirty-day check consistently, and both do it the same way: extract durable facts into a separate store, retrieve by relevance rather than recency. The rest are still spending their context window on the last forty messages.
We expect this to be the axis the category competes on next, in the way image quality was the axis two years ago.
Six platforms generate video now: OurDream AI, SweetDream AI, Candy AI, Secrets AI, Kindroid and SoulGen. A year ago the number was effectively zero.
What they produce is three to six seconds, no audio, at a resolution below their own still images, with identity drift that is visible within the clip. On the four token-metered platforms it costs several times what an image does, and a failed generation can still spend the allowance.
It is improving faster than anything else in the stack, and the gap between a 2025 clip and a 2026 clip is larger than the gap across two years of image generation. But it is currently a marketing feature, and paying more for a plan because it includes video is paying for something you will use twice.
This is the least remarked-on change and the most consistent. The demo tiers narrowed across the category this year β smaller allowances, faster walls, more features gated.
Three genuinely usable free tiers remain: Janitor AI at roughly 50 messages a day, CrushOn AI at 50 a day, and PornPen, which actually generates watermarked images for nothing. Muah AI still gives unmetered uncensored text chat. Everything else is now a demo in the strict sense β enough to see the interface, not enough to form an opinion.
Two platforms have no ongoing free tier at all: Kindroid, which can afford the position, and Seduced AI, which cannot.
Memory as the marketing axis. It is the widest quality gap, it is the one users feel most, and it is now technically tractable enough that the platforms behind will close it.
Regulation reaching the category properly. The rules on synthetic imagery of real people have tightened sharply and are still tightening, and age verification requirements are spreading. Both will change what these products can offer and where.
And consolidation. Fourteen platforms converging on one price and one feature list is not a stable arrangement, and the ones without a defensible position on either memory or image quality are the ones to be careful about committing twelve months to.
Decide which single thing you care most about, then optimise for that and accept the trade-off everywhere else. Platforms in this category are not general-purpose: the ones with the best conversation have weak image generation, and the ones with the best images forget who you are after two weeks. Trying to find one that leads in everything wastes money on subscriptions you cancel. Start with a free tier that actually lets you judge output β Janitor AI, CrushOn AI and PornPen all have real ones.
Advertised entry prices cluster between roughly $6 and $16 a month, but the sticker price is frequently not the real one. Almost every headline figure assumes annual billing; paying month to month typically costs 40β100% more. On top of that, several platforms charge a subscription for access and then bill image and video generation separately through tokens, so heavy use can double or triple the monthly cost.
Financially, almost always β the discounts are steep, often 50β65%. The risk isn't financial, it's that this category is volatile. Platforms swap the model underneath you, remove features, and occasionally close. We've watched a companion's personality change overnight on more than one platform. Committing twelve months to something with no track record is a different bet from committing to one that's been running for years, and we'd weigh that before the discount.
The flattening is real, it happens on almost every platform, and it has three separate causes that people usually blame on one. What we found across thirty-day tests, and the two platforms it does not happen to.
Read postWe signed up to all fourteen platforms without paying. Three free tiers are genuinely usable, most are demos that stop exactly where it gets interesting, and two platforms have nothing at all.
Read postEvery platform in this category calls itself uncensored and every one of them refuses things. Here is where refusals actually come from, which limits are permanent, and what the word is worth when you see it on a pricing page.
Read post
