OpenAI Jalapeño delivers higher throughput per kilowatt and lower latency than rivals on GPT-OSS 120B, with deployment set for year-end.