The AI Economy Uptime Index

A single composite number between 0 and 100 that measures the live health of the AI economy. We ping a curated set of provider control-plane endpoints every five minutes and compute the percentage of providers currently reachable. Lower than 100 means something somewhere is broken.

Every 5 minutes. A Vercel cron job hits one endpoint per tracked provider, records the latency, and writes a new index snapshot to Supabase. The /status page is cached for 5 minutes and tag-revalidated by the cron after each run.

OpenAI, Anthropic, Google Gemini, Groq, Together, Fireworks, DeepSeek, Mistral, xAI, Cohere, Cerebras, and OpenRouter. The list expands as we add more. Every provider is pinged from the same Vercel edge region so the numbers are comparable.

Three consecutive failed pings for a provider automatically opens an incident with a permanent slug. Two consecutive recovered pings auto-resolve it. Every incident gets its own citable URL at /status/incident/[slug] so researchers can reference historical outages.

No. We use the public /v1/models listing endpoint on every provider, which returns sub-kilobyte responses and does not cost anything to call unauthenticated. Some providers return 401 without auth · we treat that as reachable because it proves the control plane is up.

Yes. The Export incidents CSV button at the top of this page downloads the full archive with incident slug, provider, severity, status, start, end, duration, and peak latency. Attribution required · link back to benchgecko.ai/status.