OpenAI has unveiled GPT-5.3-Codex-Spark, a lightweight version of Codex, TechCrunch reported on Feb. 12 local time.
GPT-5.3-Codex-Spark is designed around dedicated chips to provide faster response speeds. OpenAI brought in dedicated chips from its hardware partner Cerebras and integrated them into its own infrastructure.
The chip offers speeds much faster than existing AI computation. It is designed to help OpenAI's AI models handle rapidly growing data demand.
OpenAI had previously used Nvidia H100 and custom H200 chips, but the announcement shows it is strengthening its own chip design capabilities, TechCrunch reported.
A WSE-3 chip built in cooperation with Cerebras has 4 trillion transistors and supports ultra-low-latency responses. OpenAI plans to use it to maximize real-time AI collaboration features.
Chi-gyu Hwang
delight@d-today.co.kr