How we do this
The short version: we pay for every platform ourselves, live with it for weeks, and write down what we found. No commissions, no partners, no press accounts.
Almost every comparison site in this category is an affiliate operation, and most of them rank whoever pays the most. We don't take commissions, don't run ads, and have no relationship with anything listed here β which means the only thing we have to offer you is that our opinions are actually ours. Here is how we form them, so you can decide how much weight to give them.
- 01
We pay for our own accounts
Every platform here was used on a subscription we bought ourselves. We have never asked for a press account and would turn one down β they come with better model access and higher limits than you would get, which makes the whole exercise pointless.
- 02
We use them the way you would
Not a benchmark β a habit. We build a character, talk to it for weeks, generate images, and see what happens when the novelty wears off. We deliberately check the same things on each one: does it still sound like itself after an hour, does it remember what we told it last week, how many times do we have to re-roll an image before it's usable.
- 03
We score six things, the same six every time
Six criteria, 0β10 each, weighted into one number. The weights are published below and applied identically to everything. The scores are judgements, not measurements β but they are consistent judgements, which is the most a site like this can honestly offer.
- 04
We re-check the prices
Pricing here changes constantly, and a stale price is the fastest way to waste someone's money. Every price on this site carries the date we last confirmed it on the platform's own checkout page.
Score weights
The overall score is a weighted average of the six criteria. Readers arrive here looking to switch platforms, which makes price and feature breadth matter more than they would on a pure companion-quality ranking β so value carries the most weight. The same weights are applied to every platform, and they are published here in full:
| Chat qualityCoherence over long sessions, personality consistency, and how often replies collapse into repetition. | 18% |
|---|---|
| Image qualityResolution, anatomy, consistency across generations, and how often a prompt needs re-rolling. | 20% |
| MemoryWhether the companion recalls facts from earlier sessions days and weeks later. | 12% |
| CustomizationDepth of appearance, personality, and scenario controls at creation time and afterwards. | 18% |
| ValueWhat the entry tier actually unlocks, and how fast hidden token costs stack up. | 22% |
| PrivacyData retention policy, payment discretion, export and deletion controls, and breach history. | 10% |
What we get out of this
Nothing. There are no affiliate links, no commissions, no ads and no paid placements anywhere on this site, and we have no relationship with any platform listed here. We pay for the subscriptions out of our own pockets. If that ever changes, this page will say so before anything else does.
Where this method falls short
This is one group of people's experience, not a study. Image quality is a matter of taste and ours leans a particular way. We've used these for weeks, not years, so we can't tell you how any of them behaves after six months. And the ground moves: a platform can swap the model underneath it overnight and make last month's write-up wrong. When we notice, we go back and fix it β but there will be gaps between when something changes and when we catch it.