Measured end-to-end for this short test.
Jovethra / measured evidence
A bounded gateway benchmark.
One transparent concurrency check is more useful than a generic speed claim. These results describe a short test under a specific workload, not a promise of performance.
Verified short-request checkMEDIAN LATENCY2.58 s MAXIMUM LATENCY5.22 s WALL TIME5.32 s
20 short simultaneous requests
Authorized test requests were launched within 104 ms, each capped at 16 output tokens. All 20 completed with HTTP 200.
Slowest request in the same run.
Elapsed time for the complete 20-request burst.
The result validates a bounded short-request burst only. It is not sustained throughput or a service-level guarantee, and it does not predict latency for longer prompts, streaming, tool calls, different models, or a different provider state.
Boundary observed20 succeeded 25 were rejected upstream 15 is the public global ceiling
Why the production ceiling stays lower.
All 20 short requests completed in the measured threshold test.
A separate burst of 25 short requests returned sanitized upstream errors without customer token charges.
The configured ceiling retains headroom below the observed boundary rather than marketing the maximum transient result.
Method
How to read this page.
The gateway was observed under a fixed model, short prompt, 16-token output cap, and concurrent starts. Temporary test access was removed after the run. The service status page reports current scheduled checks and rolling anonymized telemetry separately from this benchmark.
Read current status →