Skip to main content
Flatkey measures the health of every model based on real production traffic routed through the platform — not synthetic benchmarks. Health metrics are updated daily and reflect the last 30 days of actual requests.

What health metrics mean

A “success” means the upstream vendor returned a valid completion response. Errors caused by invalid prompts or context-length violations are counted separately.

Platform-wide stats

Across all models and providers, Flatkey achieves:
  • 99.9% average 30-day success rate — measured on production traffic
  • Automatic failover — when a provider endpoint degrades, Flatkey routes to a healthy alternative
  • Live dashboards — health data refreshed daily from real traffic

Top models by usage (recent)

The following models have the highest traffic volume on the Flatkey platform: This leaderboard updates daily. See the full live ranking at flatkey.ai/rankings.

How to use health data

When choosing a model for production:
  1. Prefer high success-rate models for reliability-sensitive workloads.
  2. Check average latency for real-time or streaming applications where time-to-first-token matters.
  3. Monitor your own usage in the Usage dashboard — filter by model to see your success rate.
For critical production workloads, choose models with a 30-day success rate above 99.5%. The Model Directory shows health data alongside pricing so you can optimize for both cost and reliability.

Failover behavior

Flatkey monitors upstream provider health continuously. When a provider returns elevated error rates or latency spikes:
  • Requests are automatically retried on a healthy provider endpoint.
  • Your client receives a valid response without needing to implement retry logic.
  • The failover is transparent — you see a successful response, not an error.
This mechanism underpins the 99.9% platform success rate. You benefit from it automatically without any configuration.

Viewing live health data

Real-time model health, latency rankings, and volume data are available at: