OpenAI published the first benchmark results for its custom AI inference chip, Jalapeño, on August 25, 2026, showing the chip outpacing current state-of-the-art hardware on speed and power efficiency. The record does not specify a location for the release, and no motive beyond the chip's design purpose was stated.
Jalapeño was tested on Semianalysis's InferenceX benchmark, where it delivered more tokens per user and higher throughput per kilowatt than existing leading chips, TechCrunch reported. The chip is also built to operate at scale, according to TechCrunch.
OpenAI said Jalapeño was designed specifically for AI inference workloads. The chip provides faster inference and lower latency for modern AI models while consuming less power than current leading inference hardware, OpenAI News reported.