According to gdb, OpenAI’s Jalapeño beats commercial systems on InferenceX with higher throughput per kW and lower latency, per OpenAI.