OpenAI publishes first performance results for Jalapeño custom inference chip

OpenAI released initial measured results for Jalapeño, its first custom inference accelerator. Across public models the chip delivered 1.5–1.9× more AI work per watt at peak throughput and 1.7–3.6× lower end-to-end latency than comparison systems. The results demonstrate a full-stack co-design advantage and mark the start of a multi-generation custom silicon roadmap aimed at lowering inference cost and latency.