Openai unveils codex-spark, real-time coding model with cerebras
Breakthrough in real-time coding with codex-spark
OpenAI has announced
the launch of Codex-Spark, a specialized model optimized for real-time code generation and ultra-low latency thanks to Cerebras' technology. This development follows the association revealed last month with Cerebras, in which OpenAI explained that it would progressively incorporate the company's systems into its infrastructure to support various workloads, from code generation to image creation.
Gpt-5.3-codex-spark: a powerful real-time coding model
Designed for live coding scenarios, GPT-5.3-Codex-Spark is a scaled-down version of GPT-5.3-Codex, leveraging Cerebras' Wafer Scale Engine 3 to deliver over 1,000 tokens per second while maintaining high capacity. Its performance falls between GPT-5.3-Codex and GPT-5.1-Codex-Mini.
Initial availability and future plans
Codex-Spark currently supports a 128,000 token context window and only accepts text input. OpenAI plans to expand its capabilities in the future, including support for larger models, longer contexts, and multimodal input.
Initial access and integration
Initially, Codex-Spark will be available to ChatGPT Pro subscribers, who can test the model by updating the latest versions of the Codex application, CLI, and VS Code extension. OpenAI is also offering limited API access to a select group of design partners to study how developers integrate the model into their products and services.
Combining cerebras and gpus for optimal performance
While GPUs remain the primary platform for training and inference in most cases, Cerebras' technology is better suited for Codex workloads that require extremely low latency. OpenAI notes that both systems can be combined for maximum overall performance in a single task.
Enabling instant responses and professional integration
Codex-Spark enables experimenting with instantaneous responses and facilitates the integration of low-latency models into professional environments, maintaining compatibility with popular applications and extensions like VS Code.