OpenAI has introduced a new AI coding model called GPT-5.3-Codex-Spark. The company describes it as its first real-time code writing AI model capable of writing and refining code instantly with ultra-low latency. The model runs on low-latency hardware and is specifically designed for developers. Currently, it is available in limited research preview access.
OpenAI has taken a major step in the coding domain by launching GPT-5.3-Codex-Spark, a real-time AI model built to accelerate software development workflows. The new model focuses on instant code generation and editing, signaling OpenAI’s growing emphasis on AI-powered development tools in 2026.
In 2026, OpenAI elevated the Codex model lineup to a higher strategic priority. Recently, the company released GPT-5.3-Codex, which launched ahead of the general-purpose model. OpenAI CEO Sam Altman described Codex’s growth as extremely rapid, mentioning a 50 percent increase within just one week.
With the introduction of Codex-Spark, OpenAI has now taken a direct step toward real-time coding capabilities. This move reflects the company’s stronger commitment to AI-driven software development tools.
The biggest strength of GPT-5.3-Codex-Spark is its ability to generate and edit code in real time. According to the company, the model can almost instantly write code, make edits, and reshape logic structures.
Processes approximately 1,000 tokens per second
Designed as a text-only model
Focused on code-level improvements, interface refinement, and targeted editing
Offers a 128,000-token context window
This context window enables it to handle standard and moderately complex development tasks efficiently. For more complex tasks, OpenAI recommends using the GPT-5.3-Codex model.
The high speed of Codex-Spark is powered by low-latency hardware. OpenAI recently partnered with Cerebras Systems, and Codex-Spark runs on the company’s Wafer Scale Engine 3 AI accelerator.
Internal benchmark tests suggest:
Better performance than GPT-5.1-Codex-mini
Slightly lower performance than GPT-5.3-Codex
Significant speed advantage due to hardware optimization
GPT-5.3-Codex-Spark is currently available to ChatGPT Pro subscribers via:
Codex app
CLI
IDE extension
It is being offered as a research preview to limited users. Access through API is also available to select design partners. OpenAI plans to expand access in the coming weeks.
First real-time AI coding model from OpenAI
Processes up to 1,000 tokens per second
128,000-token context window
Powered by Cerebras Wafer Scale Engine 3
Optimized for developers and software engineering workflows
Limited research preview access