AI Relay Compare

Methodology

AI Relay Compare aggregates independent probe data on 190+ AI API relay stations so you can pick one without relying on Telegram rumors. Here’s how we build the Quality Score you see on every row.

Quality Score composition

  • Model authenticity (40%) — MVP placeholder. Post-MVP, we probe each relay with fingerprint prompts and compare the responses to canonical upstream models to detect distillation, caching, or silent downgrades.
  • Latency (25%) — end-to-end probe response time. Post-MVP: measured from 5 global locations and averaged.
  • Uptime (20%) — 30-day rolling availability from daily reachability probes.
  • Price fairness (15%) — post-MVP. Compares each relay’s claimed pricing to the upstream models.dev baseline.

Data sources

  • awesome-ai-api — open-source daily probe of AI API relays. We consume the raw gateways.json and refresh it every day.
  • models.dev — structured per-model pricing and context-window data. MIT-licensed with attribution.

What’s coming

Our own authenticity probes and a price-fairness scorer are on the roadmap. Track our progress or reserve the paid plan on the Subscribe page.