Status page
Status is public. It is the first place to look when someone says “DeepSeek / MiniMax is not answering.”
Banner vs cards
- Operational — recent traffic is succeeding
- Degraded — elevated fault rate or lost host weight. Client disconnects and invalid requests do not count.
- Outage / unavailable — the model is not serving (or not listed)
- Healthy + routable on the capacity table only means shards exist. It does not mean your request will get a slot.
Numbers that matter
| Metric | Meaning |
|---|---|
| Incoming vs served | Gap is failures + 429s |
| Fault rate (5m / 1h / 24h) | Failures that count toward status. Disconnects and invalid requests have their own charts. |
| Tokens / sec | Served completion tokens, last complete minutes |
| Served share | How incoming traffic split across models |
| Load % | In-flight / effective max concurrency |
| Active shards | Funded runtimes still listed |
Epoch
Gonka rotates through inference and PoC windows. During PoC, inference capacity can dip. The page shows the current epoch stage and time left.
What to tell a user
If MiniMax/DeepSeek error rates are high and GLM is clean, say: use zai-org/GLM-5.3-Flash or retry later. Keys are rarely the issue when Status is red.