OpenAI's first custom inference chip Jalapeño beats commercial systems in efficiency

firstadopter · x · 2026-08-25

OpenAI released the first measured performance results for Jalapeño, its custom inference chip. On the InferenceX benchmark using GPT‑OSS 120B, Jalapeño delivered higher peak throughput per kilowatt and lower token latency than commercial systems. It also performed strongly on DeepSeek R1 and Kimi K2, showing cross-family adaptability. OpenAI highlights this as an example of Jevons paradox: greater efficiency expands consumption and creates new economic activity.

Related event: OpenAI's First Custom Inference Chip Jalapeño Claims Beats Nvidia Flagships(22 posts)→

Original post →

More from Infra

Infra channel →