# Methodology: how we compare Crypto LLM

Every value comes from a source of the provider, with a verbatim quote and a snapshot; agents capture it, a watcher checks the sources regularly. Scores are calculated deterministically from the data — never by a language model.

## The best service, not the cheapest

We are not a price comparison site. The score answers one question: which service is the best, judged by its features, how well it protects your privacy and how far the provider can be trusted. Price is one area among several: a fair price counts in full, only an overpriced one loses points. Whether the best service is worth its price is your call in a second step: sort and filter the list by price.

- What makes a good service weighs most. Each category places its score factors in areas with points (100 in all) and publishes every factor on its methodology page.
- Every point is a score factor bound to sourced values: a plus earns points, a minus deducts them, a note is shown without points. Each factor on the product page carries its source [s].
- Costs weigh little and count along fixed price points: a fair price scores in full, an overpriced one loses. A low price never makes up for weak privacy or trust.
- Missing data never helps: a value we have not sourced earns nothing, and what the provider does not publish counts as not met. Proofs such as an audit or a no-logs case proven in court count only when we find them.
- Hard limits: some minus factors cap the total score, however good the rest looks, for example a very young brand or a VPN that logs what its users do.
- Absolute, never relative to the best: a score changes only when the product’s own data changes; no value of another product fills a gap, and 100 does not claim perfection.

## Areas and score factors

Score = earned plus points ÷ possible plus points × 100, minus the deductions (in each area at most its points), then the caps. Possible are all plus factors that apply, also the ones not sourced yet. The value of an area (0–10) is comparable across categories, its points are not.

| Area | Points |
| --- | --- |
| Privacy | 38 |
| Security | 8 |
| Trust | 20 |
| Costs | 10 |
| Features | 19 |
| Community | 5 |

### Privacy (38 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: KYC policy | 5 | When an ID is checked: never according to the provider, not mentioned, on an occasion (suspicion, limit, payout), at any time under the terms, or for everyone before use. (highest level of: ID check, ID on demand in the terms, Sign-up needs) |
| Plus: Data at sign-up | 3 | What sign-up asks for: nothing or a username (level 0), an existing e-mail (1), a phone number or a payment method in your name (2), an identity document (3); an ID check for everyone raises it to 3. (highest level of: Sign-up needs, ID check) |
| Plus: Chats kept on the server | 8 | Not stored or kept only on your device full points, up to 30 days on the server a third, longer or until you delete them nothing; a period the policy leaves open earns nothing. (Chats stored on the server: Not stored → 100 % · Only on your device → 100 % · Up to 30 days → 33 % · Over 30 days → 0 % · Until you delete → 0 %) |
| Plus: No training on your chats | 6 | Your chats never train models full points, only until you opt out half, without opt-out nothing; silence earns nothing. (Training on your chats: Never → 100 % · Unless you opt out → 50 % · Yes → 0 %) |
| Plus: Who processes your prompts | 5 | Prompts processed on the provider's own servers full points, on rented hosts of open models half, by closed providers such as OpenAI or Anthropic nothing, routers included. (Who processes your prompts: Own servers → 100 % · Partner hosts → 50 % · Also OpenAI & co. → 0 %) |
| Plus: Confidential computing | 2 | Prompts run in a hardware enclave (TEE) that even the host cannot read. (Confidential computing (TEE): yes) |
| Plus: Sign-up over Tor/VPN | 2 | Sign-up and use work over Tor or a VPN; with an extra check there (phone, card) half. (Sign-up over Tor/VPN: Allowed → 100 % · Allowed with an extra check → 50 % · Blocked → 0 %) |
| Plus: Tor or I2P address | 1 | The official channels of the provider name a Tor (.onion) or I2P address of the service. |
| Plus: Monero accepted | 3 | Paying with Monero leaves no public trail from the payment to the account. (Pay with: XMR) |
| Plus: Cash accepted | 1 | Cash links neither a card nor a bank account to the account. (Other payment methods: Cash) |
| Plus: Legal seat | 2 | The country of the operator's legal seat by how far its law reaches user data: the same country tiers in every category (table on this page). |

### Security (8 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: Two-factor login | 5 | Two-factor login with an authenticator app and with a security key, half each; a service without accounts has no login to protect. (Two-factor login: Authenticator app, Security key) |
| Plus: Security audit | 1.5 | The provider has published an independent security audit. |
| Plus: Bug bounty programme | 1.5 | A bug bounty programme pays researchers who report flaws. |

### Trust (20 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: Operating since | 6 | Since when the current operator is on the market (first archived capture of the brand, else the founding, else the website), full after 5 years. |
| Plus: Operator named | 4 | The company behind the service is named with its registered name. |
| Plus: Imprint and terms | 3 | The provider publishes an imprint and terms of service. |
| Plus: Domain age | 3 | Age of the main domain (RDAP), full after 3 years. |
| Plus: Warrant canary or transparency report | 2 | The provider publishes a warrant canary or a transparency report. |
| Plus: Ownership disclosed | 2 | Who owns the operator is disclosed (being independent is no minus). |
| Minus: Domain bought from a previous owner | −2 | The domain had another owner before and the operator is younger than 3 years. |
| Minus: On the market for less than a year | −0 (caps the score at 5.0 when it applies) | A brand younger than a year caps the score. |
| Minus: Incidents | −8 | Serious incidents (hack, frozen funds, data leak) in the last year deduct in full, other incidents half, older ones up to three years a quarter. |

### Costs (10 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: Typical yearly cost | 10 | The cheapest way to a year of use, a subscription or 1 million output tokens a month over the API with Llama 3.3 70B or gpt-oss-120b: up to 60 USD full points, 600 USD or more none. (Yearly cost: 60 USD → 100 % · 120 USD → 80 % · 240 USD → 40 % · 360 USD → 10 % · 600 USD → 0 %) |

### Features (19 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: Frontier models | 5 | Share of the four frontier model families on offer: OpenAI, Anthropic, Google, xAI. (Frontier models: OpenAI, Anthropic, Google, xAI) |
| Plus: Open-weight models | 5 | Open-weight models (Llama, Qwen, DeepSeek …) can run without a closed provider seeing the prompts. (Open-weight models: yes) |
| Plus: OpenAI-compatible API | 4 | An API that speaks the OpenAI format works with existing tools (APIs only). (OpenAI-compatible API: yes) |
| Plus: Image generation | 1.5 | Images can be generated as well. (Image generation: yes) |
| Plus: Web search | 1 | Answers can draw on a web search. (Web search: yes) |
| Plus: Apps for web and phone | 2.5 | Apps for web, iOS and Android, a third each (chat apps only). (Apps: Web, iOS, Android) |
| Note: Uncensored models | 0 | Shown without points: whether uncensored models are a plus is still open. (Uncensored models) |
| Note: Free tier | 0 | Shown without points: the score rates the paid use. (Free tier) |

### Community (5 points)

| Factor | Points | How it counts |
| --- | --- | --- |
| Plus: Platform ratings | 5 | Average ratings on App Store, Google Play and Trustpilot, weighted by count; with too few ratings the factor is left out, never counted as zero. Single reports, Reddit and X posts do not count. |

Score version 2026-11-10.3+2026-11-10.4

## KYC: when and what

KYC is mostly your own choice, and a service with KYC can be very good. It therefore counts as a small part of privacy, the same in every category: two scales, one for when an ID is checked, one for what sign-up asks for. An ID for everyone shows in both on purpose. The badge names the policy with the level as its detail.

### KYC policy: when an ID is checked

The highest of what is sourced counts (mandatory before at any time before on an occasion); "no KYC" and "not mentioned" are told apart by what the provider states. The share is how much of the factor's points a policy earns.

| Policy | Meaning | Share |
| --- | --- | --- |
| 0 No KYC required | The provider states that it never asks for an ID, or calls itself KYC-free without publishing a limit. | 100 % |
| 1 No KYC mentioned | Sign-up asks for no ID and nothing sourced says when one would be asked. | 80 % |
| 2 Occasional KYC | An ID only on a concrete occasion: a limit, a payout, a physical card, a suspicion or a risk review. | 60 % |
| 3 KYC at any time | The terms let the provider ask anyone for an ID at any time, without naming an occasion. | 30 % |
| 4 Mandatory KYC | Everyone has to identify before using the service (the plan that can be paid with crypto). | 0 % |

### KYC level: data at sign-up

What the entry level asks for, for the score, the KYC badge and the finder. The highest requirement counts; the share is how much of the factor's points a level earns.

| Level | Sign-up asks for | Share |
| --- | --- | --- |
| 0 | No personal data | 100 % |
| 1 | E-mail or wallet only | 80 % |
| 2 | Phone or personal data | 50 % |
| 3 | ID document | 20 % |
| 4 | ID and proof of address | 10 % |
| 5 | Source of funds | 0 % |

## Trust core

The same factors with the same shares in every category, bound to facts about the provider only (not the product), so trust is comparable across the network.

| Factor | Share | How it counts |
| --- | --- | --- |
| Plus: Operating since | 30 % | Since when the current operator is on the market (first archived capture of the brand, else the founding, else the website), full after 5 years. |
| Plus: Operator named | 20 % | The company behind the service is named with its registered name. |
| Plus: Imprint and terms | 15 % | The provider publishes an imprint and terms of service. |
| Plus: Domain age | 15 % | Age of the main domain (RDAP), full after 3 years. |
| Plus: Warrant canary or transparency report | 10 % | The provider publishes a warrant canary or a transparency report. |
| Plus: Ownership disclosed | 10 % | Who owns the operator is disclosed (being independent is no minus). |
| Minus: Domain bought from a previous owner | 10 % | The domain had another owner before and the operator is younger than 3 years. |
| Minus: On the market for less than a year | 0 % | A brand younger than a year caps the score. |
| Minus: Incidents | 40 % | Serious incidents (hack, frozen funds, data leak) in the last year deduct in full, other incidents half, older ones up to three years a quarter. |

## States of a factor

- **yes**: Sourced and met: the factor earns its points (a minus deducts).
- **partly**: Sourced, partly met (a ladder step or a scale): a share of the points.
- **no**: Sourced and not met: no points, but the points stay possible.
- **not published**: The provider does not publish it: a plus counts as not met (silence never earns points), a minus deducts nothing.
- **not sourced**: Not sourced yet: earns nothing, the points stay possible and the page says “not sourced”.
- **does not apply**: Cannot apply to this product (for example no own app): left out of the possible points. Never in the KYC area.
- **not counted**: Not counted: outside the provider’s control, such as too few ratings or a price without an exchange rate.

## Provisional scores

A score is provisional (no rank, no seal) while less than 60 % of its possible points are sourced, while the KYC policy is not sourced, or while less than 50 % of the trust core is sourced.

## Value states

- **known**: The value is published and proven by a quote.
- **not published**: The provider does not publish the value (shown as such, never as 0).
- **only in the app**: The value is only visible in the provider’s app.
- **not applicable**: The attribute does not apply to this product.
- **unknown**: We have no proven value yet.

## Is the double s a typo?

No. The double s at the end of our name stands for sourced & synced. sourced: every value has a source, a quote and a snapshot. synced: our monitoring compares every value with its source continuously, and each value carries its check date.

## Sponsoring and affiliate links

Sponsored slots are labelled and have no influence on score or ranking. Affiliate links are marked.

## How we rate jurisdictions

The score factor “Legal seat” rates the country of the provider’s legal seat, i.e. the company that holds the customer contract: how far its law protects customers against account suspension, seizure and data requests. A country in no tier counts 50 %. The seat comes from the imprint, terms or company register, with its source.

| Tier | Part | Countries |
| --- | --- | --- |
| **Strong protection**: Not in an intelligence-sharing alliance and outside the EU’s cross-border orders: foreign authorities need mutual legal assistance and a domestic court order before data is handed over or a service suspended. | 100 % | Iceland, Switzerland |
| **Offshore**: Small offshore jurisdiction without an intelligence-sharing alliance: foreign requests need mutual legal assistance, but courts and transparency are weaker than in the strongest tier. | 80 % | Panama, Seychelles, British Virgin Islands, St. Kitts & Nevis, Belize |
| **EU/EEA outside the Eyes**: GDPR and court oversight, not in the Fourteen Eyes; within the EU, authorities of other member states can order data directly from providers (e-Evidence). | 60 % | Austria, Bulgaria, Croatia, Cyprus, Czechia, Estonia, Finland, Greece, Hungary, Ireland, Latvia, Lithuania, Luxembourg, Malta, Poland, Portugal, Romania, Slovakia, Slovenia, Liechtenstein |
| **Nine/Fourteen Eyes**: Member of the Nine or Fourteen Eyes intelligence-sharing alliance: working courts, but close cooperation with foreign authorities. | 40 % | Denmark, France, Netherlands, Norway, Germany, Belgium, Italy, Spain, Sweden |
| **Five Eyes**: Broad surveillance powers with gag orders, fast takedowns on copyright notices, and domain seizures by court order are common. | 20 % | United States, United Kingdom, Canada, Australia, New Zealand |
| **High risk**: State access without independent courts or a sanctions risk: authorities can block accounts and domains, sanctions can cut off payments and registries. | 0 % | Russia, Belarus, China, Hong Kong SAR China, Iran, North Korea, Syria, Cuba |

---

- **Canonical URL:** https://cryptollmss.com/en/methodology/
- **As of:** 19 Nov 2026
- **Data licence:** CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

Please cite: Source: cryptollmss.com (cryptovss), as of 19 Nov 2026, https://cryptollmss.com/en/methodology/
