Status
Generated 2026-10-27T09:00:37.960Z. Metrics contain no prompt or completion content. Uptime counts successful requests over all requests except client errors (400, 413), client disconnects, and 429, which OpenRouter also excludes. Server errors, timeouts, and interrupted streams count as failures.
Production traffic
| 24 h | 7 days | |
|---|---|---|
| Requests | 3 | 3 |
| Uptime (OpenRouter definition) | 100.00% | 100.00% |
| HTTP 429 (at capacity) | 0 | 0 |
| Time to first token p50 | 285 ms | 285 ms |
| Time to first token p95 | 285 ms | 285 ms |
| Output tokens / s per stream | 47.7 | 47.7 |
| Prompt cache hit ratio | 0.00% | 0.00% |
Benchmark traffic
Synthetic multi-turn coding workload sent through the public gateway.
| 24 h | 7 days | |
|---|---|---|
| Requests | 60 | 60 |
| Uptime (OpenRouter definition) | 100.00% | 100.00% |
| HTTP 429 (at capacity) | 0 | 0 |
| Time to first token p50 | 4416 ms | 4416 ms |
| Time to first token p95 | 25043 ms | 25043 ms |
| Output tokens / s per stream | 44.5 | 44.5 |
| Prompt cache hit ratio | 61.38% | 61.38% |
Deployment
| Region | Japan |
| Hardware | NVIDIA RTX 3090 24 GB per engine |
| Serving | vLLM 0.32, W4A16 weights, FP8 KV cache, prefix caching |
| Concurrency | 2 streams per engine; excess requests receive HTTP 429 |
Machine-readable: /status.json