OpenAI announces GPT-5.3-Codex-Spark 15x faster coding preview for Pro users
Technology
OpenAI just announced a research preview of GPT-5.3-Codex-Spark, its quickest coding AI so far: think 15-times-faster code generation and real-time results.
Right now, it's only for Pro users at $200 a month, and it runs on Cerebras's high-performance Wafer Scale Engine 3 technology.
The catch? You might hit slowdowns if lots of people are using it at once.
Codex-Spark designed for quick edits
Codex-Spark isn't here to replace the old model. It's built for those moments when you need rapid-fire code edits or quick feedback, even though it doesn't ace SWE-Bench Pro, a benchmark evaluating agentic software engineering capability.
OpenAI and Cerebras teamed up on this one, with plans to blend fast coding and deeper reasoning in future updates.