OpenAI Launches First New Model Using Cerebras Chips to Reduce Dependence on NVIDIA

OpenAI has released its first artificial intelligence model running on chips from semiconductor startup Cerebras Systems, named GPT-5.3-Codex-Spark. This is part of OpenAI's push to broaden its chip supplier base and reduce its reliance on NVIDIA. The model is positioned as a lower-power but higher-speed version of the latest Codex automated programming software. The Spark version is scheduled for release on Thursday, allowing software engineers to quickly complete tasks such as editing specific code snippets and running tests. Users can also interrupt the run at any time, or instruct the model to perform other programming-related tasks without waiting for it to complete lengthy computational processes. Last month, OpenAI signed an agreement worth over $100 billion to use Cerebras' hardware to improve the response speed of its AI models. For Cerebras, this deal provides significant leverage to compete for market share in a market long dominated by NVIDIA. For OpenAI, this collaboration is its latest move to expand its supplier base to meet growing computing power demands.