According to gdb, OpenAI’s Jalapeño beats commercial systems on InferenceX with higher throughput per kW and lower latency, per OpenAI.

According to @sama, OpenAI unveiled Jalapeño, a custom LLM chip built with Broadcom to scale ChatGPT, Codex, API, and future agentic products.

According to @gdb, OpenAI unveiled Jalapeño, an in‑house LLM inference chip with strong perf per watt, built with Broadcom for ChatGPT scale.