OpenAI's Jalapeño Chip Debuts with 1.7x Throughput per Watt Compared to GB300

OpenAI has released performance data for its Jalapeño inference chip. In tests across GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T models, the chip demonstrated 1.5 to 1.9 times higher AI throughput per watt than the GB200 and GB300. End-to-end latency was reduced to between 28% and 59% of comparative systems.