Upgrade screens in this category are deliberately fuzzy: "enhanced memory", "priority access", "premium models". Specifics would invite comparison. Strip the wording away and only four things are for sale.
1. More media: worth it, and you can do the sums
This is the most honest upgrade, since you can measure it. It means more images, more voice minutes and video access.
It is also the one case where arithmetic beats guessing. Tally what you really use, price it against buying credits on the basic plan, and go with the cheaper option. The method is in what image generation really costs you.
If the sums say the higher tier costs less, it costs less. There is nothing else to weigh.
2. Deeper memory: worth the most, hardest to check
It may be labelled longer memory or better recall. When it is real, nothing else an upgrade can buy is as valuable, because memory is what separates an app that is fun for a week from one that stays fun in month four.
The catch is proving it. "Enhanced memory" might be a bigger context window, more stored facts or better retrieval. It might also be a slogan attached to nothing at all.
The test: on the basic plan, chat daily for a week, then go quiet for four days. Come back and ask about something specific from day two, and write down what she gets wrong. Upgrade and run the identical test. If the mistakes look the same, you bought a slogan.
That is the sort of thing that takes a fortnight to pin down, and it is why we test on paid accounts over time and do not work from feature tables.
3. Faster replies: only worth it for voice
This usually means a priority queue when the servers are busy. For text, two seconds versus five is noticeable but rarely justifies a plan on its own.
Voice is another matter. Delay is the entire experience, and a three-second pause before every answer wrecks the illusion. If you are upgrading for live voice, this is the feature and not a footnote.
4. Better models: the letdown
Sold as the headline benefit, and usually the hardest to notice.
A stronger model shines at complex reasoning and long-form coherence, which is not what companion chat needs. What you actually notice here is the character description, retrieval and sampling settings, the topics in how replies are generated, and those are often identical across tiers.
It is not worthless. It is just oversold next to items one and two.
A rule that holds up
Do not upgrade until you can name the limit.
If you can say "I run out of images by Thursday" or "she forgets last week, and that matters most to me", the upgrade is a deliberate purchase aimed at a known problem.
If your reason is "the premium plan is probably better", you are buying spare capacity you cannot evaluate, and you will not know afterwards whether it helped.
Where the entry prices sit
Among the apps we track, published entry prices go from Sweetdream at US$10.00, up through Candy AI at US$12.99 and Nomi at US$15.99, to Replika at US$19.99.
A higher entry price often means fewer upgrade choices down the track, because more is bundled in from day one. Candy AI, with chat, images and voice in one subscription, is the clearest case, and it explains much of why it leads our ranking without winning any single category.
So comparing entry plans across apps can mislead. Compare the plan that includes the thing you signed up for.
The cheapest path
- Take the basic plan, monthly, for a month.
- Write down, in words, the specific limit that annoyed you.
- If it is media, do the sums. If it is memory, run the four-day test.
- Upgrade for that reason, still paying monthly.
- Think about yearly only after a month you enjoyed on the plan you settled on, following monthly, annual or credits.
Most people who follow this end up on the basic plan of a slightly dearer app and not the top plan of a cheap one, which is usually the better product for the same money.
