Status
API · api.creeksidecompute.ai
Checking readiness…
Website
If you can read this, it is up.
creeksidecompute.ai
Poe · CreeksideCompute
Public, and answers through the API above while Qwen3.8 27B is serving.
Incidents
No incidents recorded.
What "ready" means
The check calls the API's /readyz, the endpoint our own alerting watches. It answers 200 when the API can serve requests and 503 when it cannot. A 503 is either a planned pause, which this page then labels as one, or an outage. Under load a request waits up to 35 seconds for room and then gets a 429 with Retry-After, so a request refused during a busy minute is not an outage.
Measured, and dated
| Number | Value | Workload | Measured |
|---|---|---|---|
| Requests completed in full | 99.94% | All models, real traffic, 7 days; every refusal for load counted as a failure | 2026-11-09 |
| Speed, each caller | 108 tok/s | Qwen3.8 27B, NVFP4, real traffic through the public endpoint, median per request; first token 0.52 s at the median, on prompts of 12,743 tokens at the median | 2026-11-09 |
| Long prompt, first token | 34.4 s | Qwen3.8 27B, a 200,000-token prompt read from cold, through the public endpoint; 100,000 tokens in 12.9 s | 2026-11-08 |
| Context window | 1,048,576 | Qwen3.8 27B, per request; 65,536 output | 2026-10-23 |
| Speed, each caller | 98 tok/s | Gemma 4 31B, NVFP4, real traffic through the public endpoint, median per request; first token 0.69 s at the median, on prompts of 12,367 tokens at the median | 2026-11-09 |
| Long prompt, first token | 52.2 s | Gemma 4 31B, a 250,000-token prompt read from cold, through the public endpoint; 200,000 tokens in 36.0 s | 2026-11-08 |
| Context window | 262,144 | Gemma 4 31B, per request; 65,536 output | 2026-10-31 |
| Speed, each caller | 280 tok/s | Qwen3.6 35B A3B, FP8, real traffic through the public endpoint over 24 hours, median per request; first token 0.43 s at the median | 2026-11-09 |
| Time to first token | 188 ms | Qwen3.6 35B A3B, short prompt, one stream, through the public endpoint | 2026-10-14 |
| Speed, one user | 320 tok/s | Qwen3.6 35B A3B, one stream, through the public endpoint | 2026-10-14 |
| Speed, each of 16 users | 104 tok/s | Qwen3.6 35B A3B, users who pause between turns, sent straight to a server with the settings we serve; first token 0.30 s at the median | 2026-10-14 |
| Speed, each of 32 users | 68 tok/s | Qwen3.6 35B A3B, users who pause between turns, sent straight to a server with the settings we serve; first token 0.36 s at the median | 2026-10-14 |
| Context window | 1,048,576 | Qwen3.6 35B A3B, per request; 65,536 output | 2026-10-23 |
| Time to first token | 890 ms | Muse Glimmer 30B, one user in agent sessions, sent straight to a server with the settings we serve; the first token starts its reasoning | 2026-10-13 |
| Speed, one user | 63 tok/s | Muse Glimmer 30B, one user in agent sessions, sent straight to a server with the settings we serve | 2026-10-13 |
| Speed, each user at full load | 41 tok/s | Muse Glimmer 30B, users who pause between turns, sent straight to a server with the settings we serve; first token 1.40 s at the median | 2026-10-14 |
| Context window | 131,072 | Muse Glimmer 30B, per request; 65,536 output. A 114,886-token prompt answered correctly in 17.9 s | 2026-10-14 |
Every figure on the site comes from a run, or a read of our request records, that we keep; ask for the raw runs and we send them. API reference · Privacy · Content policy