Skip to content

Gemini AI Gateway Benchmark: Gemini (Direct)

Gemini (Direct)

Request Latency

Baseline
99% success

No gateway involved — a direct call to Gemini's API, used as the no-gateway control every other provider is compared against.

Cold E2E
1863ms
Median
Warm TTFT
2016ms
Median
Tokens/sec
101
Median
DNS
0.6ms
Median
TCP
1.9ms
Median
TLS
5.3ms
Median

Cold Connection Breakdown

Where a fresh request's time goes: DNS lookup, TCP connect, TLS handshake, then time to first response byte, then first byte to first streamed token.

DNS0.6ms
TCP1.9ms
TLS5.3ms
To First Byte1845.4ms
First Byte → Token0.0ms

Performance Over Time

Iteration Distribution

Cold E2E

Warm TTFT

Tokens/sec

DNS

TCP

TLS

Methodology

Every run executes on a GitHub Actions runner provided by Namespace, physically located in Northern Virginia, US. Since results are a property of the network they're measured from, treat rankings as relative to that vantage point rather than a universal number — see Limitations in the archived methodology.

What We Measure

Every gateway is hit directly with the same prompt and model (Gemini 3.6 Flash), addressed the way each API expects it to be addressed — no gateway is proxied through another. Each request is either cold (a brand-new TCP+TLS connection, opened from scratch) or warm (a second request reusing an already-open keep-alive connection).

Cold requests are broken into DNS, TCP, and TLS phases (real socket timestamps, not estimates), plus time to first response byte and first streamed token. Every request also records output tokens/sec and success/failure.

This family's no-gateway control is a direct call to Gemini's own API, not a competing participant — it isolates how much latency each gateway adds on top of the underlying model provider. See the other target-model families on the AI Gateway Benchmarks overview. Full methodology, including how "cold" and "warm" are defined and their limitations, is documented in the archived benchmark methodology.

How We Score

The Composite Score combines cold end-to-end latency and warm time-to-first-token — the short-lived-process case and the steady-state case — with generation throughput, all multiplied by success rate.

  • • Cold E2E median: 30%
  • • Warm TTFT median: 30%
  • • Cold E2E P95: 15%
  • • Warm TTFT P95: 15%
  • • Tokens/sec: 10%