New Lightweight Codex Model
OpenAI on Thursday announced a lightweight version of its agentic coding tool Codex, releasing GPT-5.3-Codex-Spark as a smaller model designed for faster inference. The company said Spark is built for speed and marks a deeper integration with Cerebras, its hardware partner, by using a dedicated chip to power the model.
Cerebras Partnership And Infrastructure
The Cerebras partnership was announced last month, when OpenAI said it had reached a multi-year agreement with the company valued at more than $10 billion. At the time, OpenAI said bringing Cerebras into its compute mix was aimed at making its systems respond faster. OpenAI now described Spark as the first milestone in that relationship and said the new model runs on Cerebras’ Wafer Scale Engine 3, the company’s third-generation waferscale chip that contains 4 trillion transistors.
Product Positioning And Availability
OpenAI said Spark is designed for swift, real-time collaboration and rapid iteration, and positioned it as a tool for daily productivity and quick prototyping rather than longer, heavier tasks handled by the standard 5.3 model. The company said Spark is focused on the lowest possible latency in Codex and is intended to operate alongside a mode for longer-running work. Spark is available in a research preview for ChatGPT Pro users in the Codex app.
Statements From OpenAI
In its announcement, OpenAI said Codex-Spark is meant to support two complementary ways of working, with real-time collaboration for fast iteration and longer tasks for deeper reasoning and execution. The company added that Cerebras’ chips are suited to workflows that require extremely low latency.
Altman’s Pre-Announcement Post
Before the release, OpenAI chief executive Sam Altman posted on X that a special update would be coming to Codex users on the Pro plan and said it “sparks joy for me,” a message that appeared to point to the Spark launch.
Cerebras Background And Funding
Cerebras has operated for more than a decade and has taken on a more visible role in the AI sector in recent years. The company said last week that it raised $1 billion in new funding at a valuation of $23 billion and has previously said it intends to pursue an initial public offering.
Cerebras View On The Release
Sean Lie, the company’s chief technology officer and co-founder, said the focus of GPT-5.3-Codex-Spark is on working with OpenAI and developers to see what fast inference enables. He said the goal is to identify new interaction patterns, new use cases, and a different model experience, and described the preview as an initial step.
Featured image credits: Canopy Creative
For more stories like it, click the +Follow button at the top of this page to follow us.