techBreaking

OpenAI Unveils Lightning-Fast GPT-5.3-Codex-Spark for Real-Time Coding

OpenAI has launched GPT-5.3-Codex-Spark, a breakthrough real-time coding model capable of generating code at over 1,000 tokens per second—marking the company's first major deployment on non-Nvidia hardware through a partnership with chip maker Cerebras.

By Cody RodeoUpdated Feb 15, 2026 • 4:18 AM

The new model, available as a research preview to ChatGPT Pro users, delivers 15x faster code generation compared to previous versions while maintaining high capability for real-world coding tasks. Running on Cerebras' Wafer Scale Engine 3—a purpose-built accelerator for low-latency inference—Codex-Spark transforms how developers interact with AI coding assistants.

"This is designed as a daily productivity driver, helping users with rapid prototyping," explained OpenAI in the announcement. The model features a 128k context window and excels at quick coding tasks rather than longer, more complex projects.

In demonstration videos, OpenAI showed Codex-Spark building a snake game in seconds, showcasing the dramatic speed improvement over standard models. The near-instant code generation enables developers to iterate rapidly and see results in real-time as they type.

The Cerebras partnership represents a strategic shift for OpenAI, which has historically relied exclusively on Nvidia's GPUs. Industry analysts view this as OpenAI hedging its bets on hardware suppliers and exploring specialized chips optimized for specific AI workloads.

The launch comes as competition in AI-assisted coding intensifies, with companies racing to deliver faster, more capable development tools. OpenAI's move toward ultra-low latency positions Codex-Spark as a direct competitor to GitHub Copilot and other real-time coding assistants.