OpenAI releases GPT-5.3-Codex-Spark, a smaller AI encoding model that generates over 1,000 tokens per second on Cerebras hardware. It is OpenAI’s first GPT model that does not run on Nvidia.