Activity
Measured, not claimed
Every figure here is read from the gateway's own request log. Nothing on this page is an estimate.
Model status — today
Some models degraded today
deepseek-v4-flash
96.1% answered · 5,029 today
Degraded
24h agoNow
deepseek-v4-pro
99.9% answered · 3,996 today
Operational
24h agoNow
deepseek-v4.1-flash
96.9% answered · 17,315 today
Degraded
24h agoNow
glm-5.1
100% answered · 37 today
Operational
24h agoNow
glm-5.2
99.8% answered · 909 today
Operational
24h agoNow
glm-5.3
99.7% answered · 12,760 today
Operational
24h agoNow
glm-5.3-flash
99.7% answered · 3,383 today
Operational
24h agoNow
glm-5.3-flashx
99.7% answered · 5,331 today
Operational
24h agoNow
hy3-tencent
90.9% answered · 11 today
Degraded
24h agoNow
hy4-preview
97.7% answered · 184 today
Degraded
24h agoNow
kimi-k2.7
96.8% answered · 95 today
Degraded
24h agoNow
kimi-k3
98.4% answered · 3,381 today
Operational
24h agoNow
minimax-m3
100% answered · 21 today
Operational
24h agoNow
Health per model over the last 24 hours, from the gateway's own request log. Operational at 98%+ of requests answered, degraded at 90–98%, down below 90%. Requests a client cancelled itself are not counted against a model.
- 5.06M
- Requests served, all time
- 60.56B
- Tokens, last 7 days
- 97.9%
- Requests answered, last 7 days
Detail
- Requests, last 24 hours
- 92,569
- Requests, all time
- 5,061,939
- Tokens, last 7 days
- 60,558,017,843
- Models live
- 13
How these are counted
- Requests answered
- Calls that returned an answer, as a share of all calls in the last seven days. Requests a client cancelled part-way are left out of both sides, since a cancelled call is not a failure of the gateway.
- Tokens
- Prompt and completion tokens added together, across every model, for the last seven days.
- Refresh
- Counters are read from the live database and held for five minutes, so a reload inside that window shows the same figures.
Response time is not shown here because it is set by the upstream provider and varies by model, so a single number would misrepresent it.