AI Provider Latency Tester
Ping major AI provider endpoints from your browser and compare response times live.
Network requests only fire when you click. No keys, no payloads — timing only.
Select providers and press Run test. You'll get a time-to-first-byte bar per provider, measured from this browser. Expect CORS-blocked responses — the timing is still reported as an approximation, labeled honestly.
How it works
This tester sends lightweight requests from your browser to the public API hostnames of the providers you tick — Anthropic, OpenAI, Google, Mistral and Groq — and times how long each takes to answer. Nothing runs until you press the button: every probe is explicitly user-initiated, and no API keys or payloads are involved. The measurement is the delta between two performance.now() calls wrapped around a fetch, which captures DNS resolution, TLS handshake and the network round-trip to the provider edge.
An honest caveat sits at the center of the design: browsers enforce CORS, and AI providers do not let arbitrary web pages read responses from their API origins. When the browser blocks a response, the request has still traversed the entire network path — the server answered, the browser merely refused to expose the body. We therefore catch the error and report the time-to-error as an approximation of round-trip latency, clearly labeled as such. It is not a perfect TTFB, but it is a consistent, comparable signal across providers when measured the same way in the same session.
What this tool cannot tell you is how fast a model generates tokens. Edge latency and inference throughput are different layers: a provider can answer your TCP connection in 40 ms and still take eight seconds to stream a long completion. Treat these bars as a connectivity check — useful when you suspect an outage, a regional routing problem, or your own network being the bottleneck — and pair them with the official status pages linked from our Provider Status tool when something looks wrong.
Results vary run to run because the public internet varies: congestion, cold DNS caches and Wi-Fi conditions all move the numbers. Run the test two or three times and read the relative ordering rather than any single absolute figure. If you are offline, the tool says so plainly instead of showing fake data. For teams that need this signal continuously rather than on-demand, FORG records per-request latency from real agent traffic across every session.
Frequently asked questions
What exactly does this tool measure?
It measures the wall-clock time from your browser issuing a request to a provider's public API hostname until the first response (or error) arrives, using performance.now(). That covers DNS, TLS, and network round-trip to the provider's edge — it is a connectivity and edge-latency probe, not a model inference benchmark. Model generation speed is a separate question; use our Streaming Latency tool for that.
Why do some probes show 'timing to CORS error'?
Browsers enforce cross-origin rules, and AI API endpoints do not allow arbitrary websites to read their responses. The request still travels the full network path and the server still answers — the browser just refuses to hand us the body. The time until that refusal is a usable approximation of round-trip latency, so we report it and label it honestly rather than hiding it.
Does my network connection affect these numbers?
Yes, heavily. Every measurement includes your local network, your ISP, and your geographic distance to the provider's nearest edge. A slow result here can mean your Wi-Fi is congested, not that the provider is down. Run the test a few times, compare relative differences between providers rather than absolute numbers, and cross-check the provider's official status page before concluding anything.
Why does the tool wait for me to click 'Run test'?
Privacy and predictability. FORG tools never make network calls behind your back — pages that fire requests on load are both a privacy smell and a source of noisy measurements. You decide when probes run, you can re-run them on demand, and nothing about your visit is sent to any provider until you click. The rest of the page works completely offline.
Is this the same as checking a provider's status page?
No — it is complementary. Status pages report what the provider's own monitoring sees, often with a delay of several minutes during incidents. This tool tells you what your machine sees right now, which catches regional routing issues and ISP-level problems that a global status page never will. Our Provider Status tool aggregates the official incident history side.
Built by FORG — AI cost observability for agentic coding. Free tools, no signup, nothing leaves your browser.
Learn about FORG