The 100-point method
Every app is scored out of 100 across five weighted dimensions. The rubric is identical for all apps, so editorial judgments use a consistent framework. These scores are not measured benchmark results or proof of a statistically significant difference between apps.
What can we verify—and what remains unverified?
Source-audit version: October 12, 2026. Each row has its own check date. This is a comparison of public evidence, not a controlled hands-on performance ranking. Unknowns stay unknown; editorial scores are unchanged.
| App / source check | Model & platform | First limiter & full bill | Memory proof | Exit & privacy | Provenance |
|---|---|---|---|---|---|
| Kindroid2026-10-31 | Flagship preview; Ember trial; Polaris/legacy after paid conversionWeb, iOS, Android documented | One-time 50 messages (15 some regions); then read-onlyStandard + optional Ultra + optional MAX + extra media/voice; checkout price varies | Cascaded Memory and journals documented; comparative accuracy not benchmarked | Read-only retains view/playback/export/delete; add-ons cancel separatelyLegal explicitly says not E2E; automated review and bounded authorized access; UK/EEA/Swiss erasure timing has retention exceptions | Official subscription/legal/FAQ/purchase document APIs; no paid test; Oct 29 recorded prices separatePrimary source ↗ |
| Nomi AI2026-10-24 | Exact live model/version not verifiedUS Apple listing; other checkout routes unverified | Free account documented; live cap not account-testedUS iOS: $15.99 Standard (confirm cadence) / $39.99 quarterly / $99.99 yearly; separate credit packs | Continuity is an editorial preference, not a controlled cross-app result | Cancel through payment channel; downgrade keeps one-to-one chat while paid group/voice/video becomes read-only; content preservation not guaranteedAbout 28-day personal-data deletion wording retains legal/archive/de-identified training-data caveats | US Apple listing, beginner guide, cancellation/refund and privacy checked October 24; no account recall/checkout testPrimary source ↗ |
| Candy AI2026-10-31 | Exact model/version not public in reviewed termsOfficial web terms | Terms generally cap free users at 5 messagesSubscription + monthly token amount at checkout (generally 100) + optional refills | Durable memory and correction controls not benchmarked | Cancel in settings; terms say token expiry, help says hidden until resubscribing; conflict unresolvedTerms/privacy disclosures are not an independent architecture audit | Official terms/privacy and public Help API; no paid test; Oct 29 AU price record separatePrimary source ↗ |
| Character.AI2026-10-12 (age); 2026-09-16 (plans) | Proprietary model; exact version not verifiedWeb/mobile verification controls documented | Standard free chat; premium allowance/ads vary; age classification is separatePreviously documented US Lite $4.99 / C.AI+ $9.99 monthly; current checkout and Charms extra | Paid features do not establish superior long-term recall or remove filters | Cancel through payment channel; deletion is separate; export recovery not testedC.AI names Persona for selfies/ID fallback and says IDs kept one week; processor policy recheck blocked | Official public Help API; subscription evidence separately datedPrimary source ↗ |
| Janitor AI / DeepSeek2026-10-12 | deepseek-flash (V4.1) / deepseek-v4-pro; old Flash aliases remappedCustom API route; Janitor+ Router documented as web-only | Provider balance/rate limit; Router also needs Janitor+Direct provider tokens OR Janitor+ + Router wallet; these are separate balances | Model context is not proof of persistent personal memory | Provider keys and balances managed separately; no tested migration receiptEach connected service has its own policy; do not send a key to an unknown proxy | Provider docs and Router help; legacy Janitor proxy article 404; no end-to-end account testPrimary source ↗ |
| OurDream AI2026-10-18 | Exact model/version not verifiedOriginal ourdream.ai web product/policies | Free signup; lasting message/media cap unverified$19.99 monthly OR $119.88 upfront yearly; 1,000 monthly dreamcoins + optional refills | Marketing claim; no multi-session recall/correction test | Cancel in account settings/support; period-end access; media retention untestedHomepage E2E claim not independently audited; policy documents collection and retention exceptions | Original homepage, refund/privacy policies; no paid benchmarkPrimary source ↗ |
| FantasyGF2026-10-17 | Exact chat/media model version not verifiedOfficial web terms | Free signup; lasting allowance unverifiedTerms list $12.99 monthly / $29.99 quarterly / $69.99 yearly + media credits | Long-term recall/correction not benchmarked | Cancel in account settings/support; terms describe deletion/anonymization with exceptionsPublic policy, not an independent zero-access guarantee | Official terms; live paid account limits untestedPrimary source ↗ |
| DreamGF2026-10-17 | Exact model/version not verifiedOfficial web membership policy | Free cap unverified; paid membership + base 100 monthly tokensCrossed monthly/quarterly renewal labels; base 100 tokens: image 10, companion 20, voice 3, calls 7/min | Persistent recall and editing not demonstrated | Ordinary cancellation: period-end access. Bonus offer: 30-day lock. Used membership restricts refunds; possible $1.99/month maintenance feeNo independently verified architecture or universal retention deadline | Official pricing/membership/terms/privacy; checkout, charge applicability and quality untestedPrimary source ↗ |
| Dondi AI2026-10-12 | Exact model/version not verifiedOfficial homepage and privacy policy | Free signup promoted; account cap/price unknownSubscription/media total not independently verified | Homepage claim; no persistent recall receipt | Policy describes rights; actual export/deletion/cancellation controls untestedHomepage zero-log/E2E claims not audited; policy describes logs, content, retention and possible model training | Homepage-policy cross-check; no paid benchmarkPrimary source ↗ |
A future hands-on comparison must record the same tasks, dates, model/version, platform, free/paid state, recall and correction checks, costs and exit steps for every app. A vendor claim alone does not pass that protocol. Read our full method.
Which five dimensions do we score?
Weights reflect what actually matters in a long-term companion. Privacy is weighted heavily on purpose: these are intimate conversations, and 2025–2026 brought real regulatory action across the industry, so data handling is not a footnote.
| Dimension | Weight | What it measures |
|---|---|---|
| Conversation quality | 25 | Naturalness, coherence and emotional awareness of the dialogue. |
| Memory & consistency | 25 | Long-term recall and whether the companion stays the same "person" over time. |
| Privacy & data | 20 | Encryption, export, deletion, and whether chats train the underlying model. |
| Customization | 15 | Control over personality, voice, appearance and backstory. |
| Value | 15 | True monthly cost once add-ons and tokens are included. |
| Total | 100 | The published match score for each app. |
How do we gather evidence?
Our assessments are research-based, not hands-on testing. We never claim to have "used an app for 14 days." Instead, scores are built from:
- Official product information and feature documentation
- Public pricing pages (and the real cost of add-ons/tokens)
- Documented company history and regulatory filings (e.g. FTC, the Italian Garante)
- Synthesized feedback from across user communities
- Reporting from outlets including TIME, STAT and Fortune
How often do we update our reviews?
Pricing and policies in this category change often, so we review and refresh ranked pages roughly every 30 days. AI answer engines and search both favour recently updated pages — and more importantly, stale prices would make the rankings wrong.
How do we maintain editorial independence?
Some outbound links are affiliate links (marked
sponsored nofollow), and we may earn a commission if you
subscribe. Commercial relationships may influence a clearly labelled
Featured Partner or CTA placement, but never the 100-point evidence score
or editorial rank. Featured units display the app’s real score and rank.
See our
affiliate disclosure and
editorial policy.
What are the current scores?
| App | 25 | 25 | 20 | 15 | 15 | Total |
|---|---|---|---|---|---|---|
| Nomi AI | 22 | 23 | 16 | 13 | 13 | 87 |
| Kindroid | 20 | 21 | 18 | 15 | 10 | 84 |
| Candy AI | 21 | 17 | 13 | 13 | 12 | 76 |
| Replika | 17 | 15 | 12 | 14 | 13 | 71 |
| Character.AI | 20 | 11 | 15 | 10 | 13 | 69 |
| Pi | 21 | 16 | 14 | 5 | 12 | 68 |
| SpicyChat AI | 19 | 12 | 11 | 13 | 12 | 67 |
| Janitor AI | 19 | 12 | 10 | 14 | 11 | 66 |
| CrushOn AI | 18 | 12 | 11 | 12 | 11 | 64 |
| PolyBuzz | 18 | 11 | 12 | 11 | 11 | 63 |
| Chai | 18 | 11 | 11 | 11 | 11 | 62 |
| Talkie AI | 17 | 11 | 9 | 12 | 12 | 61 |
| Anima AI | 16 | 12 | 12 | 10 | 10 | 60 |
| Kupid AI | 18 | 11 | 10 | 11 | 9 | 59 |
| HeraHaven | 16 | 10 | 10 | 13 | 9 | 58 |
| FantasyGF | 15 | 10 | 10 | 13 | 9 | 57 |
| GirlfriendGPT | 16 | 10 | 9 | 12 | 9 | 56 |
| uDesire AI | 15 | 10 | 9 | 12 | 9 | 55 |
| DreamGF | 14 | 9 | 9 | 14 | 8 | 54 |
| Muah AI | 15 | 10 | 8 | 11 | 9 | 53 |
| Paradot | 16 | 16 | 10 | 8 | 3 | 53 |
| Dream Companion | 15 | 12 | 8 | 12 | 5 | 52 |
| Joi AI | 15 | 9 | 9 | 11 | 7 | 51 |
| ChatFAI | 15 | 8 | 8 | 10 | 9 | 50 |
| Our Dream AI | 15 | 12 | 9 | 9 | 4 | 49 |
| Dondi AI | 14 | 12 | 8 | 9 | 5 | 48 |
| Fanfinity | 14 | 9 | 7 | 11 | 6 | 47 |
| Go Love AI | 13 | 8 | 7 | 11 | 7 | 46 |
| Darlink AI | 13 | 8 | 7 | 10 | 7 | 45 |
| Dreamz AI | 12 | 8 | 8 | 10 | 6 | 44 |
| DreamBF AI | 12 | 7 | 7 | 10 | 7 | 43 |
Column headers show each dimension’s maximum weight. Scores reflect the latest publicly listed information we could verify; always confirm current details on each app’s own site.