What is codex-spark?

Codex-Spark is OpenAI’s real-time coding AI model that generates code at over 1,000 tokens per second. That’s fast enough that it feels instant when you’re typing. It was released in February 2026 as a speed-optimized version of GPT-5.3-Codex, designed specifically for live coding where you need immediate feedback.

What makes Codex-Spark different is the hardware it runs on. Instead of Nvidia GPUs like most AI models, it uses the Cerebras Wafer-Scale Engine, a specialized chip that’s optimized for extreme speed. This is OpenAI’s first model not running on Nvidia hardware.

The model is designed for iterative, real-time work. It has a 128k context window, which means it can keep track of a lot of code at once, and it’s fast enough that you can use it in the flow of actually writing code rather than waiting for responses. It’s available through ChatGPT Pro, the Codex command-line tool, and a VS Code extension.

Codex-Spark is available as a research preview for ChatGPT Pro subscribers at $200 per month. The speed makes it practical for tasks where you want the AI to assist as you type, not just generate large blocks of code and move on.

I haven’t tried Codex-Spark yet. At my day job I use Claude Code with Opus, which is already pretty fast, but I’m curious how the instant feedback of 1,000 tokens per second would change the experience.