OpenAI has introduced GPT-5.3-Codex-Spark, a new AI model specifically optimized for programming. Unlike previous versions, it doesn’t run on Nvidia GPUs but on wafer-scale chips from Cerebras Systems, meaning a notable strategic shift. The model delivers more than 1,000 tokens per second, making it about 15 times faster than its predecessor.
The main goal is ultra-responsive, real-time, interactive code development. For now, Spark is available as a research preview for ChatGPT Pro users and selected partners. The clear focus is on speed rather than maximum general intelligence. With a 128,000-token context window and text-only output, OpenAI also signals a broader hardware strategy aimed at further reducing latency.