GPT-5.3-Codex-Spark: Ultra-fast Model for Real-time Coding

Hey developers! Get ready for a serious speed boost in your coding workflow. OpenAI just dropped a game-changer: GPT-5.3-Codex-Spark. This isn't just another model; it's an ultra-fast, real-time coding assistant designed to keep up with your thoughts.

What it's for

GPT-5.3-Codex-Spark, introduced on February 12, 2026, is a smaller, highly optimized version of GPT-5.3-Codex. Its main gig? Supercharging real-time coding within the Codex app. Think instant suggestions, rapid refactoring, and getting work done in the moment. It's built for those quick, interactive coding sessions where you need immediate feedback. At launch, it's text-only but boasts a generous 128k context window, ready to understand your larger codebases.

Why it matters

Speed is the name of the game here. Codex-Spark is engineered to deliver over 1000 tokens per second, making it feel almost instantaneous. This isn't magic; it's the first major fruit of OpenAI's partnership with Cerebras, running on their incredible Wafer Scale Engine 3 AI accelerator. We're talking near-instant inference on ultra-low latency hardware!

And it's not just fast; it's smart. On benchmarks like SWE-Bench Pro and Terminal-Bench 2.0, Codex-Spark demonstrates strong performance, completing tasks in a fraction of the time compared to its larger sibling, GPT-5.3-Codex. Plus, OpenAI has really honed the entire request-response pipeline. You'll experience an 80% reduction in client/server roundtrip overhead and a 50% faster time-to-first-token, all thanks to some clever optimizations and persistent WebSocket connections. This means a smoother, more responsive coding experience across the board!

Where you get it

Ready to give it a spin? GPT-5.3-Codex-Spark is currently available as a research preview. If you're a ChatGPT Pro user, you can dive in via the Codex app, CLI, and VS Code extension. For select design partners, it's also accessible through the API. Keep in mind that during this research preview, its usage is governed by separate rate limits, so it won't impact your standard usage.

This is just the beginning for OpenAI's foray into ultra-fast models. Codex-Spark is the first of many! Future plans include even larger models, longer context lengths, and eventually, multimodal input. Keep an eye on the OpenAI Homepage for updates!

Try it: Introducing GPT-5.3-Codex-Spark for the official announcement and to start your real-time coding journey!