OpenAI retires GPT-5.3-Codex-Spark as usage declines

Raw speed cannot substitute for deep reasoning in coding tasks, and ultra-fast niche models are losing ground to versatile frontier models that have become both fast and capable. Developers prioritize accurate code generation and complex problem-solving over raw latency once generation speeds reach an acceptable threshold. The rapid deprecation of experimental preview models demonstrates how quickly developer expectations shift, while retiring low-usage models allows OpenAI to reclaim compute resources and simplify model routing.