Performance Statistics — Cortega
Performance

One gateway. 52,000 requests per second.

One 8 vCPU instance sustained 52,000 req/s at 100% success, adding a few milliseconds per call. Cortega’s data plane is agentgateway; on identical hardware it carried about 36× the throughput of LiteLLM.

Request a demoDownload — free
One benchmark. Three numbers. Same hardware. SUSTAINED THROUGHPUT — REQUESTS/SEC (HIGHER IS BETTER) Cortega data plane ~36,000 req/s LiteLLM (rust mode) 983 req/s p99 LATENCY ADDED — MS (LOWER IS BETTER) Cortega data plane ~2 ms LiteLLM (rust mode) 71 ms MEMORY UNDER LOAD Cortega data plane ~25 MB LiteLLM (rust mode) 2.15 GB · ~85× more agentgateway 1.4.0 vs LiteLLM 1.98.0 (rust mode). Fortio, mock backend, 32 connections. Source: agentgateway.dev, August 2026.
The numbers

The numbers

52,000requests per second on one 8 vCPU gateway, at 100% success
~2 msmedian gateway overhead at 40,000 requests per second
~$0.002per million requests for the gateway node at full load
36×the throughput of LiteLLM’s Rust mode, same hardware (agentgateway.dev)
Head to head

LiteLLM: 983 req/s

agentgateway benchmarked against LiteLLM on identical hardware: Fortio load generator, mock backend, 32 connections, June and August 2026. LiteLLM’s Rust mode measured slower than its Python mode.

GatewayRequests / secp99 latency addedMemory
Cortega data plane (agentgateway)~36,000~2 ms~25 MB
LiteLLM, Rust mode98371 ms2.15 GB
LiteLLM, Python mode3,19832 ms11.8 GB

LiteLLM publishes a 0.66 ms p99 figure for its Rust gateway. Benchmarked independently on that build, it sustained 983 requests per second at 71 ms p99.

Results

The results

Load generator, gateway, and control plane on separate nodes. Upstream is a mock server with a fixed 60 ms delay, so the only variable is what the gateway adds. Each row is 30 seconds at a fixed rate.

vCPU$/hrRequests/secEnd-to-end p50Gateway overheadSuccess
2$0.0875,00061.7 ms~2 ms100%
2$0.0878,00061.9 ms~2 ms100%
2$0.08712,00064.8 ms~5 ms100%
4$0.17415,00060.6 ms~1 ms100%
4$0.17425,00061.9 ms~2 ms100%
8$0.34840,00062.0 ms~2 ms100%
8$0.34848,00062.7 ms~3 ms100%
8$0.34852,00064.9 ms~5 ms100%
8$0.34853,00069.3 msout of headroom99.97%
Reading the numbers

What it means

01

Scales with cores

A 4 vCPU node held 25,000 req/s. Doubling to 8 vCPUs roughly doubled that, to more than 52,000, still at 100% success.

02

Overhead stays low

About 2 ms up to 40,000 req/s, rising to about 5 ms near the limit. Streaming passes through token by token with no added buffering.

03

One gateway is the floor

Cortega runs as a fleet. Capacity grows as you add gateways, at the same cost per request, placed close to where traffic originates.

04

About $0.002 per million

The 8 vCPU node costs about 35 cents an hour. Counting the whole test rig, it stays under a cent per million requests.

Get started

Bring your own benchmark.

Try it yourself

Foundation is free — one gateway, standard guardrails, no time limit.

Download — self-hostedSign up — managed cloud

Talk to us

Tell us what you’re evaluating and we’ll run Cortega against it on your terms.

Request a demoContact sales