The fastest model in OpenAI's history is about to be discontinued. Tibo, the project lead for Codex, announced that the GPT-5.3-Codex-Spark model will be retired next week. Capable of processing 1,200 Tokens per second, this model is OpenAI's first to break free from dependence on NVIDIA and the first deliverable from the 750-megawatt Cerebras deal. However, from its release to discontinuation, it only survived for 7 months.