Explore

Looking for our schedule?

Next benchmark run

private

AI Gateway — Anthropic daily

Sun, Sep 6 · 9:00 AM

Benchmark summaries

Latest results

Next benchmark Friday5d 18h 32m 36s

Sandboxes

Composite score · Higher is better

Composite score · Higher is better

Composite score · Higher is better

Composite score · Higher is better

Cloudflare AI Gateway

What We Benchmark

Five suites covering the infrastructure AI products actually run on

Time to interactive

Sandboxes

Cold-start latency and reliability across sandbox providers, measured in a 100-sandbox burst.

View benchmarks

Clone → install → typecheck

Real-world workload

A full dax-style developer loop inside a fresh sandbox — system packages, Bun, a pinned opencode clone, install and typecheck — timed phase by phase.

View benchmarks

Latency and throughput

Storage

Upload and download of a 10 MB object against each provider's bucket, with each operation timed independently.

View benchmarks

Actions per second

Browsers

Per-action throughput inside a live headless session — 10 sequential actions across 100 sessions, in stealth mode at 1920×1080.

View benchmarks

Cold E2E and TTFT

AI gateways

Connection-phase latency, time to first token, and generation throughput for each gateway, scored against a direct-to-Anthropic baseline.

View benchmarks

Methodology

How the Numbers Are Made

Every benchmark is reproducible: open code, fixed workloads, results published in full

One workload, every provider

The same script, payload, and model hit every provider, addressed the way its own API expects. Nobody gets a tuned path.

Repeated on a schedule

Every suite runs many iterations per provider, automated on a recurring schedule, so results reflect steady state rather than one lucky request.

Scored on outcomes, not one lucky run

Every family has its own scoring function, but none reward a single fast run. Latency is blended from median and tail percentiles, completion is counted in phases, and coverage or quality is measured where those matter. All scored families multiply by success rate so providers that fail often lose real points.