Status page
How to read the public status page: what the red percentage counts, and what it leaves out.
On this page
openbroker.gonka.gg/status is the public view of recent inference traffic. It refreshes from a cached snapshot about once a minute. It does not show host addresses.
Model cards
Each card is a model family (DeepSeek V4 Flash, MiniMax M2.7, GLM 5.3 Flash). Kimi is omitted because that model is deprecated.
The percentage is the share of recent requests that failed for a reason other than capacity.
Counted in the percentage:
- HTTP 502 timeouts, including
timeout awaiting response headers - Other HTTP 4xx and 5xx that are not capacity
- Upstream and invalid-request failures
Not counted in the percentage:
- HTTP 429
- Messages about rate limits, no available host, escrow capacity, or a participant request budget
Those are demand. They appear in the reason breakdown as Over limit and Out of capacity. A card can look healthier than the share of calls that returned a completion, if many calls were rejected for capacity.
Which window the badge uses
| Badge | Rule |
|---|---|
| Operational | Error rate at or below 8% |
| Elevated errors | Error rate above 8% |
| High error rate | Error rate above 25% |
The rate is the last 5 minutes when that window has at least 5 requests. Otherwise it is the last hour. If the hour also has fewer than 5 requests, it is the 24 hour rate. A quiet model can therefore show a badge from an older window.
The page-level line (all models operational, some models elevated, multiple models disrupted) summarizes those cards.
Served versus incoming
Incoming is requests seen. Served on the model card is HTTP successes (status under 400).
The green line on the traffic chart is drawn from requests that were not counted as errors. Because capacity rejections are not errors, that line can be higher than the number of completed responses. Use the card's served count when you want completions.
Charts cover 24 hours, 7 days, and 30 days. Reason colors on the breakdown are timeout, upstream error, client disconnect, invalid request, over limit, and out of capacity.
What a red page does not tell you
A high percentage means a large share of non-capacity calls failed. It does not by itself say your key is invalid. Check one failed row in usage: a 401 or 402 is your account, a 502 with a header timeout is the wait described in Streaming, and a 429 is concurrency.
Use the same page to decide whether to send less or more. That procedure is in How much to send.