OpenAI reports first Jalapeño accelerator benchmarks

OpenAI reported first benchmark results for its Jalapeño AI accelerator at a media briefing, saying the chip delivered 1.5 to 1.9 times more AI work per watt at peak throughput and 1.7 to 3.6 times lower end-to-end latency than comparable systems listed on InferenceX. The company said the tests used single-token prediction without speculative decoding. Jalapeño is designed around close memory and compute affinity to reduce data movement. OpenAI plans small-volume rollouts at year-end and a larger ramp in 2027.