How fast are AI API outage alerts from llmlatency.dev?

Providers are probed from 4 regions every five minutes; an outage is confirmed when a failure is seen from two or more regions at once with no other provider failing, so an alert can follow within one probe cycle of confirmation.

Is the free alerts tier really free?

Yes. The measured data is published under CC BY 4.0 and is already available as RSS and JSON; the free tier is the same data as a daily e-mail digest.

What does a founding subscription include?

Real-time alerts by e-mail, Slack or webhook, per-provider and per-region filters and 90-day history export, at $19 per month frozen for the first 20 seats; nothing is charged until delivery is live.

Why not rely on the provider's status page?

A status page shows what the provider chose to admit after its own thresholds; these alerts come from independent multi-region measurement, and the incident log shows both records side by side.

Alerts: know when an AI API fails or a model is retired

Updated September 06, 2026 · 45 providers · 4 regions

The same measured record that powers this site — outages confirmed from 4 regions, the model deprecation calendar, price changes — delivered to you, instead of you finding out from a failing request or a status page that never admitted anything.

Where this stands, September 2026: RSS and JSON are live; e-mail and webhook delivery is being built.

This form is the waiting list and our measure of demand. Leave an address, pick the plan you would actually use, and we will write once when delivery is live. No card is asked for, nothing is charged, and no automated confirmation e-mail is sent yet.

What is there to be alerted about right now?

Confirmed outages, 72 h

failed from 2+ regions at once, nobody else failing — log

Model shutdowns ahead

scheduled by providers within the next 90 days — calendar

Probe cycle

5 min

an alert can follow a confirmed failure within one cycle

Plans — pick one, leave an address

Free$0Daily digest by e-mail: confirmed outages, new deprecation notices, price changes — the same data as the RSS and JSON feeds, which work today.Founding · first 20 seats$19/month, price frozenReal-time: e-mail, Slack or webhook within minutes of a confirmed multi-region failure; deprecation notices the day they appear; per-provider and per-region filters; 90-day history export. Reserve a seat now — nothing is charged until delivery is live and you confirm.Businessfrom $199/monthA feed for your product: JSON and webhooks with commercial re-use, an SLA, custom slices (your providers, your regions). Pick it here, or write to the address below the form.

Optional: what to send, which providers, what for

What to send Confirmed outages Model deprecations Price changes Providers that matter to you What would you use this for?

One e-mail when it launches; an unsubscribe link in every message after that. Your address is stored on our own server, not handed to a mailing-list vendor. No tracking pixels, no third-party scripts, and the form works with JavaScript switched off. Business enquiries: contact@llmlatency.dev.

How fast would an outage alert be?

Every provider is probed from 4 regions every five minutes. A failure is confirmed when it is seen from two or more regions at once and no other provider is failing in the same minutes — the same rule as the incident log. An alert can therefore follow within one probe cycle of the second region confirming; a single-region failure is never sent as an outage.

Why not just watch the provider’s status page?

Because a status page shows what the provider decided to admit, after its own thresholds. Only 10 of the 45 providers tracked here publish a machine-readable status feed at all. These alerts come from independent measurement, and the incident log shows both records side by side, so you can see which one noticed first.

What counts as a deprecation alert?

A new entry on a provider’s own deprecation page: the model, the announcement date, the shutdown date, the documented replacement and the notice period in days. Every entry carries the source URL it was verified against; nothing is inferred from blog posts or rumours.

Already available without signing up

RSS — confirmed outages and reports as they are published
incidents.json · deprecations.json · rankings.json — CC BY 4.0
MCP server — the same data as tools for AI agents

Method and limitations: methodology · Corrections: contact@llmlatency.dev