OpenAI's Jalapeño Chip Outperforms Current AI Hardware on Key Inference Benchmarks

OpenAI has developed a custom chip called Jalapeño, designed to handle fast AI inference at scale. The chip was evaluated using Semianalysis's InferenceX benchmark, a standard tool for measuring AI hardware performance. Results showed Jalapeño delivered more tokens per user than existing state-of-the-art chips. It also demonstrated superior throughput per kilowatt, indicating stronger energy efficiency. The benchmarks suggest Jalapeño could represent a significant step forward in AI inference hardware.