Reference · Glossary
P95 latency
The response time below which **95% of requests** finish — catches slow tail better than average alone.
When to use
SLOs and alerts for chat and RAG APIs.
When not to
As the only metric — pair with error rate and quality proxies.
Example
p95 = 1.8s while mean = 0.9s → a few slow retrieves or long generations need traces.