OpenTela

Mesh observatory

Real throughput measured at the gateway while serving live LLM inference — time to first token and output tokens per second per GPU model and served model, aggregated over the trailing window. No benchmarks, no self-reported numbers: just the traffic the mesh actually answered.