OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show

Jalapeño outperformed the current state of the art on Semianalysis’s InferenceX benchmark, delivering both more tokens per user and higher throughput per kilowatt. The result suggests improved inference efficiency (better compute utilization per energy cost) relative to existing models. The article provides performance-directional upside but no magnitude in absolute terms, limiting near-term market impact.