NEUTRAL SCORING PROTOCOL

How We Score AI API Gateways

Every provider is scored 0–100 across five equally-weighted dimensions. Calculations are strictly empirical. No score is paid for, sponsored, or influenced by providers. We maintain a strict editorial firewall.

[01] SCORING DIMENSIONS SPECIFICATIONS

[DIM-01]

Transparency

20 pts

Does the provider have a public about page? Are model names and prices clearly disclosed? Is the pricing model (per-token vs. subscription) unambiguous? Providers that hide credit-to-token conversion rates lose points here.

// DIAGNOSTIC CRITERIA MATRIX

Full pricing page with per-model rates[20 PTS]
Subscription tiers without token conversion[10 PTS]
No pricing information without signup[0 PTS]
[DIM-02]

Support Quality

20 pts

Can you reach a human when something breaks? Email and ticket-based support score highest. Telegram-only is weak but better than nothing. WeChat-only is a red flag for non-Chinese users.

// DIAGNOSTIC CRITERIA MATRIX

Email + GitHub issue tracker[20 PTS]
Telegram community + DM[12 PTS]
WeChat QR only[5 PTS]
[DIM-03]

Payment Safety

20 pts

Stripe payments protect users with chargeback rights. Crypto-only or WeChat-only payments offer no recourse if the provider goes offline. We check which processors are accepted.

// DIAGNOSTIC CRITERIA MATRIX

Stripe only[20 PTS]
Stripe + crypto[16 PTS]
Alipay/WeChat only[8 PTS]
Payment method unknown[0 PTS]
[DIM-04]

US/EU payment ease (board column)

Display only

Separate from the trust-score payment dimension: the Pay column grades how easy it is for a typical US or EU buyer to check out without Alipay, WeChat, G2G credits, or crypto-only rails. A = Stripe/Paddle-class checkout; F = CN-wallet-only or contact-to-buy.

// DIAGNOSTIC CRITERIA MATRIX

A — Stripe / Paddle[Easy PTS]
B — Card or Stripe + CN option[OK PTS]
C — Card + SBP / YooMoney / crypto[Friction PTS]
D — Crypto or marketplace credits[Hard PTS]
F — WeChat/Alipay-only or unknown[Avoid PTS]
[DIM-05]

Community

20 pts

Is there a visible developer community around this service? GitHub repos, Discord servers, real IDE integrations (Claude Code, Cursor, Cline), and social presence all signal that real developers have tried and vouched for the product.

// DIAGNOSTIC CRITERIA MATRIX

GitHub org + Discord + 100K users claimed[20 PTS]
Discord community only[12 PTS]
No social presence[0 PTS]
[DIM-06]

Longevity

20 pts

How long has the provider been operating? New domains (especially cheap TLDs like .me, .top, .cloud) score lower. Providers with Wayback Machine history, established customer bases, or stated founding dates score higher.

// DIAGNOSTIC CRITERIA MATRIX

Founded 2024 or earlier, verifiable history[20 PTS]
Founded 2025, some community evidence[12 PTS]
Founded 2026, no archive history[5 PTS]

[02] TRUST SHIELD CLASSIFICATION TIERS

Trusted75–100 PTS

Independent verification, strong developer community, Stripe/Paddle payouts, and 100% transparent pricing.

Verify First50–74 PTS

Verifiable uptime but check token details, potential China-adjacent structures, or thin documentation logs.

Caution25–49 PTS

Critical trust gaps. Extremely new domains, unverified support responses, or completely opaque pricing conversion.

Avoid0–24 PTS

Immediate visual threat indicators. Chinese WeChat/Alipay-only rails, fake custom model naming, or high fraud alerts.

[03] HEURISTIC DIAGNOSTIC THREAT FLAGS

In addition to the core 0–100 trust matrix, we apply alert tags to highlight infrastructure risks, pricing anomalies, or data routing paths.

[ALERT]

China-Adjacent Operator

Providers operated from China or using Chinese payment methods (Alipay, WeChat Pay) create data residency concerns. Your API calls — including prompts, code, and documents — may pass through infrastructure subject to Chinese law. This is a risk signal, not an automatic disqualification.

[ALERT]

More Expensive Than Official

Some gateways claim to be discount services but price certain models above official Anthropic/OpenAI/Google rates. This can happen due to currency conversion, profit margin on specific models, or simple pricing errors. We flag this prominently.

[ALERT]

Custom Model Naming

Some China-adjacent providers use non-standard model names (e.g., gpt-5.2, gpt-5.4) that do not correspond to real OpenAI products. This makes it impossible to verify what model is actually being called.

[ALERT]

Credit Systems Without Token Conversion

Subscription services that sell 'credits' without disclosing how many credits equal 1M tokens make price comparison impossible. We flag these providers and exclude them from per-token price comparisons.

Scoring Protocol Constraints & Limitations
  • Latency metrics rely on continuous p50/p95 regional telemetry probes, but real-world performance depends heavily on custom route workloads.
  • Per-token pricing grids are checked hourly but final calculations depend entirely on vendor credit conversion rates.
  • "China-adjacent" flags represent regulatory and prompt data residency transparency alerts, not product stability rankings.
  • Verifiable compliance claims (e.g. SOC2) represent scraping validation states, not audited legal audits.