Default Nginx proxy buffering spikes Claude SSE streaming TTFT from 0.7s to 3.4s
Tool · DEV Community · stat: 3.4s p50 Nginx default proxy buffering settings silently convert Server-Sent Events streams into chunked batch jobs, delaying Claude API responses. Benchmarks across 200…
Tool · DEV Community · stat: 3.4s p50
Nginx default proxy buffering settings silently convert Server-Sent Events streams into chunked batch jobs, delaying Claude API responses. Benchmarks across 200 prompts show default reverse-proxy buffering spiked time-to-first-token p50 latency from 0.71 seconds to 3.4 seconds, solvable via proxy_buffering off or X-Accel-Buffering: no headers.
Legacy reverse proxy defaults are silently destroying real-time LLM streaming latency. Founders building AI chat interfaces must audit proxy buffering configuration to eliminate multi-second latency spikes on streamed token responses.
Every claim ties to a primary source. See our methodology.