llmlatency.dev
An independent third-party project monitoring the availability and latency of AI inference APIs, continuously probing the reachability of major model providers' APIs from data centers in multiple regions.
Timeline (last 90 days)
The independent monitoring project llmlatency.dev continuously probed 46 AI inference APIs over a 30-day window from September 18 to October 17, 2026, collecting data from four regions—Germany, US Central, Tokyo, and São Paulo—with one request every 5 minutes, totaling 1,580,667 requests.
The rule for determining an attributed outage is: a failure is attributed to the provider itself only when the same provider fails simultaneously from 2 or more probe regions and no other provider shows a similar failure in the same period.
Network-layer availability is not equivalent to inference success; a case where a host responds but the model queue is stuck still counts as "online" under this monitoring methodology.
All Facts (3)
The independent monitoring project llmlatency.dev continuously probed 46 AI inference APIs over a 30-day window from September 18 to October 17, 2026, collecting data from four regions—Germany, US Central, Tokyo, and São Paulo—with one request every 5 minutes, totaling 1,580,667 requests.
50%UnverifiedThe rule for determining an attributed outage is: a failure is attributed to the provider itself only when the same provider fails simultaneously from 2 or more probe regions and no other provider shows a similar failure in the same period.
50%UnverifiedNetwork-layer availability is not equivalent to inference success; a case where a host responds but the model queue is stuck still counts as "online" under this monitoring methodology.
50%