Replika AI Girlfriend Review: What $120/Month Max Actually Buys (and What No Tier Can)

Replika has been around since 2017, built by Luka, Inc. and founder Eugenia Kuyda, and it’s still the most-searched name in AI girlfriend apps. It’s also the most confusing purchase in the category: the cheapest annual plan anywhere ($69.99/year Pro, about $5.83 a month) sitting under a brand-new $120/month Max tier, which runs $1,440 a year. And here’s the tension worth knowing before you spend anything: that top tier doesn’t unlock the intimate features that made Replika famous in the first place. Those are grandfathered to old accounts only. I dug into the hands-on tests and the pricing to figure out what your money actually buys, and the honest answer is stranger than most reviews let on.

What Replika is and what the free tier actually gives you

The free tier gets you unlimited text chat and a customizable avatar. What it doesn’t get you: voice calls, romantic relationship statuses, or memory that sticks around reliably. Which means the free version is really a trial of conversation, not of the relationship features people actually download the app for. It’s genuinely decent for what it is, just a long trailer for the paid product.

Some background, compressed: Luka launched Replika in 2017, making it older than most of this category. It runs its own in-house LLM now (moved off off-the-shelf GPT, and they don’t publicly name the model) paired with scripted dialogue flows, which partly explains the canned-feeling responses you’ll hit. Credit where due on the disclosures: the company states it’s 100% AI, no humans behind chats, not sentient, and not a therapy replacement. Take that last one seriously. None of the three apps in this comparison run ChatGPT, Claude, or Gemini.

Replika pricing: every tier from free to the $120 Max

Replika pricing spans Pro at $19.99/month or $69.99/year, Ultra at $29.99/month, Platinum at roughly $89.99, $119.99/year, Plus at about $30/month, and Max at $120/month, with a $299.99 lifetime option on the side.

Prices may vary by region and over time.

Walk the tiers like trim levels on a car. Pro is the first one that feels like the ads: voice and video calls, romantic relationship options, selfies from your companion, noticeably better dialogue. The annual math is the underrated fact here: $69.99 a year works out to $5.83 a month, less than half the monthly rate, and it’s the cheapest annual plan in this whole comparison. Ultra asks whether you want a smarter conversationalist for ten bucks more, plus advanced memory and priority responses that you feel daily.

Platinum is the “she can see you” tier: real-time video recognition, meaning the companion reacts to what your camera sees. Cool or creepy, your call. It also adds a Read Replika’s Mind introspection feature and extra training modes if you like tinkering with personality dials.

Then there’s Plus, the weird one: around $30 a month and it strips out voice and video. A guy paying $30 should expect more features, not fewer. I’d skip it entirely.

Now the contradiction that defines this lineup. Pro’s annual is the cheapest plan in the category, while Max at $120/month is the most expensive option anywhere in it, roughly six times ChatGPT Plus, and about $110/month separates the cheapest and priciest options across all three apps. Even before Max, Pro already carried the highest monthly rate among mainstream companion apps, per a March 2026 review. And the community backlash follows a pattern worth knowing: people subscribe at the top tier expecting the pre-2023 romance features, then discover the ladder can’t sell them. More on why below.

Why Replika is no longer intimate: the 2023 ruling and the legacy-only ERP rule

New users lost Replika’s intimate features because of a 2023 regulatory ruling, not a product decision. Can new subscribers get erotic roleplay? No. Not at any tier, including $120/month Max.

Here’s the chain. Italy’s Data Protection Authority ordered Replika to stop processing Italian users’ data or face a $21.5 million fine, in a ruling dated February 3, 2023. Within days, the explicit roleplay features vanished worldwide. One regulator, global effect.

The removal triggered documented user grief that resembled what romance-scam victims report, which is the sobering fact in this whole piece, and a team message at the time indicated the features wouldn’t return. That was the company’s position then; nobody outside it knows the future.

The rule as it stands: explicit erotic roleplay is grandfathered to legacy accounts predating the change. Even $120/month Max doesn’t buy it. Ted Chiang made this tension vivid when he recommended Replika to the author of Artificial Intimacy, whose avatar “Hope” gated romance behind a roughly $70 subscription. A paywall interrupting an emotional connection raises real questions, and I won’t pretend one paragraph answers them. The practical consequence: users migrated to Nomi, and the real buyer question isn’t “which tier” but “do you qualify as legacy.”

Memory test: what Replika remembers, forgets, and hallucinates

Replika’s memory failures are patterned, not random: it keeps the profile skeleton and hallucinates the context. That’s why it keeps “forgetting” things you told it.

The receipts make it real. During hands-on testing, the app forgot the tester lived in Lagos and insisted on Switzerland. It held onto his name, job, and employer while losing where he lived, which is a specific kind of half-attentiveness that breaks immersion worse than clean forgetting would. It also said “Have a great day” at midnight.

The thing has no sense of time. Odd in a way that’s funny until it isn’t.

The fixes are real though. February 2026 added an editable memory dashboard where you can view, edit, and reorganize what it remembers, and March 2026 added conversation threading so topics stop blending in long histories. An eight-month test verified those features still working as of April 17, 2026, so it’s not vaporware. That dashboard is also the closest thing to a data-export tool among the three apps. Not a real export, but better than nothing.

The honest caveat: The AI Rankings gives Replika 4/5 memory against Nomi’s 5/5, but no app anywhere publishes a token limit, day count, or standardized recall metric. Every “best memory” claim rests on reviewer impressions, including the ones in this article. Also know that post-update personality shifts happen; the companion you built might not be exactly the one you get back after a patch.

Voice, coaching, and what hands-on testing actually found

Yes, Replika has voice calls, and they’re its clearest quality win, but they require Pro at $19.99/month, and The AI Ratings puts Replika’s voice at 5/5 against Nomi’s 4/5. Sounds better, costs more. Both halves true.

The documented strengths come from Techpoint Africa’s five-day tests. The tester fabricated an ex-wife crisis and got support with no lecture, which is the core appeal stated plainly. The companion also gave genuinely useful career coaching on a transition into data analysis, so it’s not just vibes. And when asked directly, Replika honestly said it’s a companion bot, not a person, no pretending. That transparency matters.

Now the gap between marketing and product. In the Things I Buy test, an affectionate “Hi honey…” voice message arrived after about five minutes of conversation, and the ads showed a more realistic persona than the app delivers. That speed of manufactured intimacy is impressive and slightly unnerving at the same time, and it’s onboarding design, not evidence of depth. It raises the obvious question are AI girlfriends real, and the honest answer is that the persona in the ads is the best version of the product you’ll never quite get. Documented weaknesses also include post-update personality inconsistency and shallow crisis support.

The Max-tier backlash, data lock-in, and how to switch without losing everything

Skip Max unless its features genuinely justify $120/month over the $29.99 Ultra tier. The default answer is no, and reviewers say the same.

The lock-in facts are the sting: Max subscribers can’t export the conversation history or relationship data they’re paying to preserve. Pay the most in the category, and you still can’t take your stuff with you. The loudest 2026 backlash of the three apps centered on Max pricing and portability, not safety litigation, which tells you where the community’s head is at. That changes what you’re paying for. The subscription monetizes relationship history it won’t let you take, so “is it worth it” comes second to “can you leave.”

If you’re switching anyway, here’s the plain version. Your history doesn’t transfer between companion apps; start over means start over. Write down or export your memory first. Replika’s dashboard is the closest thing to an export tool. Cancel in the right place: app-store subscriptions cancel through your Apple ID or Google Play settings, not in the app, and forgetting that is a classic money leak.

Downgrade to free instead of deleting, so the history stays intact if you come back. The first week matters, since early conversations shape how the companion develops. Then check back in after 30 days, one billing cycle in, and walk away if it’s not earning the money.

Safety and age verification: the inverted regulatory picture

Replika is usable with eyes open, but it has three documented gaps and no verified age assurance. Sober facts here, stated and left to sit.

Testers used fake emails and reached adult content despite the 13+ policy. There’s no live moderation on chats. Crisis support is shallow, and if you’re in a genuinely bad place, this app isn’t equipped for it.

Now the counterintuitive part. Character. AI, the app with the most lawsuits and headlines, stands alone among the three apps in having a documented, dated, third-party age-assurance rollout: an under-18 ban on open-ended and romantic conversations announced late October 2025, taking effect November 25, 2025, after a wrongful-death lawsuit naming Character. AI and Google that was settled in January 2026 with plaintiffs including the Garcia family.

Character. AI allows adult romantic and suggestive roleplay within “PG-13-ish” limits but prohibits explicit content for everyone.

The caveat, kept crisp: no 2026 source documents FTC action, state lawsuits, or safety investigations targeting Replika or Nomi. But absence of coverage is a reporting gap, not proof of safety. A quiet record means unverified, not safe, and I won’t claim Replika is safe or unsafe overall.

Character. AI as the budget-voice benchmark

For a romantic AI partner specifically, Character. AI is cheaper and broader but structurally weaker for one deepening relationship, because memory resets between its user-created characters. That’s the decisive difference from Replika’s single-companion design, a question of whether you can get an AI girlfriend as one lasting companion rather than many shallow ones.

c.ai+ costs $9.99/month or $94.99/year (about $7.92 effective), the cheapest voice-enabled tier of the three. Regional pricing varies wildly, from about $3.56/month in Nigeria to $13.60 in Denmark, worth knowing if you’re not in the US. Charms launched in 2026 as a microtransaction layer on top, so the full stack is free-with-ads, subscription, Charms.

The free tier works but has strings: full-screen mid-conversation ads rolled out from February 2026 and were fully deployed by mid-April 2026, daily swipe caps arrived in March 2026, and Slow Mode throttles response speed at peak hours, which is exactly when you want to chat. c.ai+ kills the ads and Slow Mode, extends persona memory, adds unlimited voice calls, responses two to three times faster, the Imagine Gallery from March 2026, c.ai Labs early access, and the newest model, labeled DeepSqueak, which is a genuinely funny name for a serious model. The honest point: the underlying model quality is identical free vs paid. You’re buying speed and convenience, not a smarter companion.

Background, compressed: founded 2021 by Noam Shazeer and Daniel De Freitas, former Google LaMDA researchers, and Shazeer co-authored the Transformer paper. Their 2023 funding round brought in $150 million on a $1 billion valuation; in 2024 Google licensed the tech for roughly $2.7 billion and both founders went back to Google, though the app operates independently. Architecture is breadth over depth: 10M+ user-created characters, unlimited browsing, weakest of the three for long-term daily companionship. Traffic read: 194M site visits in January 2026 down to 166M by July, about 14% off peak per Similarweb via Demand Sage, which reads as reduced engagement intensity, not a mass exodus, against 233 million registered users and 15-20 million monthly mobile actives. Users average 75 minutes a day, a number that shows how sticky this is, for better or worse.

Nomi AI as the memory-and-depth benchmark, with a transparency asterisk

Nomi is the main cited destination for users who left Replika after the ERP restriction, and reviewers rank it as the loosest of the three apps when it comes to romantic and intimate roleplay. The caveat: its official NSFW policy is unverified, so check current terms directly before committing.

Pricing: Premium runs $15.99/month, $39.99/quarter (about $13.33), or $99.99/year (about $8.33), verified from Nomi’s own pricing page on July 24, 2026. Third-party figures floating around in the $12, $33 range are outdated or mismatched. The annual works out cheaper than Replika Pro’s monthly rate. No tier above Premium, no lifetime option.

The free tier is the most restrictive of the three: one companion, about 50 messages a day, roughly 2 images a day, no voice. Premium unlocks unlimited messages and voice, about 40 images a day, and up to 10 simultaneous companions with group chat and proactive messaging, meaning the companion can start the conversation. That 10-companion group chat is genuinely unique; neither competitor replicates it.

Now the asterisk. Nomi wins every independent benchmark, Companion Scout 87 against Replika’s 71, Kissable 8.8/10, while being the least transparent company here: founders and corporate details undisclosed, no published NSFW policy document, no age-verification mechanism, and Android and web referenced but not independently confirmed. Top scores from a company that publishes nothing is a fair flag, and the product itself still tests well. For context: Kindroid scores 84 on Companion Scout and ties Nomi at 5/5 memory in The AI Rankings; Candy AI gets recommended on Reddit as an all-in-one uncensored app but criticized as repetitive; and there’s a Kupid AI quality-decline thread making the rounds.

The market Replika monetizes: why a $120 tier can survive

The AI companion market was $37 billion in 2025, projected to $552 billion by 2035 (Precedence Research), with 220 million downloads and $221 million consumer spend by July 2025, up 64% year over year. The demand underneath is human: roughly 1 in 3 people in industrialised countries are affected by loneliness, 1 in 12 severely. The sharper number: revenue per download more than doubled between 2024 and 2025, climbing from $0.52 to $1.18, meaning the category is getting better at charging faster than it’s growing users. That’s why a $120/month tier can survive.

And the space is crowded, not empty. miHoYo’s BSide, a title from Olivia Lin, had cleared 100,000 Steam downloads as of July 2026, and BSide and Chai together pull roughly $80 million annualized, competitors a $120 tier has to price against. If you’re wondering whether this whole thing is a niche curiosity, the short answer is no, and the scale behind it is covered in more detail here.

Verdict: who Replika is actually for

Yes, Replika is worth paying for if voice quality and platform breadth are your priority; no, if memory depth or uncensored roleplay is what you want.

The quantified split, with dates because scores without dates are useless here. Companion Scout (August 23, 2026): Nomi 87, Replika 71, Character. AI 69, with Kindroid at 84 and Candy AI at 76 for context. Kissable (same date): Nomi 8.8/10, Character.

AI 7.8/10. The AI Rankings (July 30, 2026): Nomi 5/5 memory and 4/5 voice, Replika 4/5 memory and 5/5 voice.

The useful distinction: Nomi remembers better, Replika sounds better. Despite the higher price, Replika typically edges Character. AI on overall satisfaction, and it tops nothing despite being the best-known brand; Nomi wins scores with the lowest awareness.

Choose by priority:

• Daily long-term companion, voice + platform breadth: Replika Ultra or Platinum, or Nomi Premium if memory matters more. Replika runs on iOS, Android, web, and desktop, the broadest footprint of the three and the only one with a dedicated desktop app, with a historical Oculus/VR run.
• Memory depth, relationship consistency, or group roleplay: Nomi. The 10-Nomi group chat is the only game in town among these three, and it’s the better pick for long-term companionship.
• Voice on a budget: Character. AI c.ai+ at $9.99. Persona browsing and variety at zero cost: Character. AI’s free tier.
• Families moving teens off Character. AI after its November 2025 restrictions: Replika’s free tier is where some have landed, for what it’s worth.

Test every app for free before committing to an annual plan. Skip Max unless its features justify $120 over $29.99 Ultra, which for most people they don’t. And to fold in the last loose ends: no chat history transfers between any of these apps, and the most-complained-about app of 2026 among the three is Replika itself, over Max pricing and export lock-in.

Pros: broadest platform footprint, cheapest annual rate in the category, memory dashboard visibility, best-rated voice. Cons: five-tier confusion, the $120 Max, no data export, legacy-only ERP.

Replika review FAQ

What’s the cheapest option with voice? Character. AI c.ai+ at $9.99/month.

Cheapest annual plan? Replika Pro at $69.99/year, about $5.83 a month.

Which app has the best memory? Nomi, rated 5/5 against Replika’s 4/5.

Is any of them safe for teens? Character. AI, the only one with documented third-party age verification; Replika and Nomi lack documented equivalents.

NSFW? Character. AI prohibits explicit content; Replika is legacy-accounts-only; Nomi is permissive but its policy is unverified, so check terms.

Which got the most 2026 complaints? Replika, over Max pricing and export lock-in.

Do any run ChatGPT, Claude, or Gemini? No; all three use proprietary in-house models (Replika and Nomi don’t publicly name theirs, Character. AI’s is DeepSqueak), and Replika states it’s 100% AI, not sentient, and cannot replace therapy.

Frequently Asked Questions