Free tools

LLM API latency & status tester

Test OpenAI, Anthropic and Gemini side by side with your own API keys: is the API answering, how long until the first token, and how long the whole reply takes.

  • Free
  • No sign-up
  • Results in seconds

How the LLM API latency & status tester works

  1. 1Paste an API key for OpenAI, Anthropic, Gemini or all three
  2. 2We stream a short prompt to each model from our servers
  3. 3Compare time to first token, total time and speed

What it checks

  • Time to first token (TTFT)
  • Total response time
  • Tokens per second
  • API key and model access
  • Rate limits and overloads
  • Median of 3 runs
Example result, with sample data.

Frequently asked questions

What is time to first token (TTFT)?
The time from sending a request to receiving the first piece of the model's reply. With streaming, it's how long your users stare at an empty box before text starts to appear, so it matters more for chat than total time does.
Is it safe to paste my API key?
Your key is sent to our server over HTTPS, used for the test requests and then dropped: it isn't stored, logged or added to the page address. If you'd rather not use your main key, create a separate key with a low spending limit for testing, and revoke it afterwards.
How much does a test cost?
Each run is one short request: a one-line prompt and a reply of under 100 tokens, a tiny fraction of a cent on small models. Reasoning models also bill the tokens they spend thinking, so they cost a little more.
Why is the first run slower?
The first request also opens the connection to the provider (DNS, TCP and TLS). Later runs reuse it, like a real app does. With 3 runs we show the median, so one slow run doesn't skew the result.
Why is a reasoning model's first token so slow?
Models like gpt-5 and Gemini 2.5 think before they answer, and the thinking isn't streamed as visible text. TTFT here is the time to the first visible token, so it includes the thinking time your users would wait through.
What do 401, 429 and 529 mean?
401 or 403: the API key was rejected or doesn't have access to that model. 429: you hit a rate limit or ran out of credits. 529 (Anthropic) or 503: the provider is overloaded, which is an outage on their side rather than a problem with your key.
Where do the requests come from?
From our servers, not your browser, so the numbers reflect the provider's API and a data-center network rather than your own connection. Your app's latency also depends on where it runs.
Can I check my LLM API around the clock?
Yes. Spot Downtime's LLM API monitoring sends a short prompt to your model on a schedule and alerts you by email, Slack, SMS or a phone call when it stops replying, returns errors, or gets slower than a limit you set.

Get told the moment it breaks

Checking by hand only tells you about right now. Spot Downtime checks your sites around the clock and alerts you by email, Slack, SMS or a phone call.

Start monitoring free