ENGINEERING STUDY

P95 Matters: The Hidden Tail Latency of AI Providers

Median latency hides the requests your users actually complain about.

Key findings
  • 12 of 13 providers (92%) show p95 TTFT above 2× their median.
  • The widest tail in the pool: Nebius at 3.7× (p50 288 ms → p95 1061 ms).
  • The tightest: Cortecs at 1.9× — proof that a narrow tail is an engineering choice, not luck.

The ratio table

p95/p50 is the honest tail signal: it is scale-free, so a fast and a slow provider can be compared directly on how much they degrade.

Providerp50 TTFTp95 TTFTRatio
Nebius288 ms1061 ms3.68×
Together401 ms1452 ms3.62×
OVHcloud372 ms1345 ms3.62×
Nextbit341 ms1158 ms3.40×
Cerebras148 ms501 ms3.39×
SiliconFlow587 ms1986 ms3.38×
Mistral298 ms963 ms3.23×
Groq179 ms530 ms2.96×
Fireworks312 ms918 ms2.94×
OpenRouter384 ms1106 ms2.88×
Scaleway356 ms1025 ms2.88×
DeepInfra512 ms1337 ms2.61×
Cortecs525 ms1003 ms1.91×
TTFT tail ratios by provider (p95 / p50) · DEV DATA
What this means for engineers
  • For interactive workloads, set SLOs on p95, not p50 — and pick providers with ratios near the pool's best.
  • A low median with a wide tail often indicates aggressive batching; verify under your own concurrency.
Dataset

Published figures are pinned to this snapshot and are not recalculated retrospectively. View the current benchmark →

Cite this researchInferenceBench (2026). P95 Matters: The Hidden Tail Latency of AI Providers. InferenceBench Research.
← All research
Related research