When OpenAI introduced GPT-5.3-Codex-Spark, the headline takeaway was obvious: real-time coding, near-instant responses, and more than 1,000 tokens per second. But beneath the performance metrics sits ...