OpenAI Unveils GPT-5.3-Codex-Spark on Cerebras Chips: A Game-Changing Shift from Nvidia for AI Giants

OpenAI has unveiled GPT-5.3-Codex-Spark, its first AI model utilizing Cerebras Systems’ hardware, moving away from its reliance on Nvidia. This new model is designed for efficient, low-power coding tasks, enhancing interactive development workflows by focusing on specific code edits and targeted tests. The ChatGPT Pro subscribers will initially get access as a research preview. Optimized for minimal edits and high throughput, it can exceed 1,000 tokens per second with the right setup. Cerebras’ Wafer Scale Engine, featuring hundreds of thousands of AI cores, minimizes data movement, thus reducing latency in inference workloads. While OpenAI continues to depend on Nvidia for its main training infrastructure, this partnership marks a strategic expansion into a dedicated low-latency compute tier. The Codex model has over 1 million active users and will soon be available beyond Pro users as performance evaluations continue. For more tech insights, follow Tom’s Hardware for the latest news and reviews.

Source link