OpenAI Debuts Faster, Leaner Codex Model on Custom Cerebras Hardware

OpenAI unveiled a streamlined version of its Codex programming tool on Thursday, engineered for speed. The new model, GPT-5.3-Codex-Spark, is a lighter variant of the full Codex platform released earlier this month, built specifically for rapid response times. A key component of this launch is the specialized computing hardware powering it: the Wafer Scale Engine 3 chip from OpenAI's partner, Cerebras.

This move represents the first tangible product from a multi-year, $10 billion partnership between the two firms announced last month. OpenAI stated that incorporating Cerebras technology aims to significantly accelerate how its AI systems respond. The WSE-3 chip, Cerebras's third-generation design, packs four trillion transistors onto a single wafer to handle the computational load.

Described as a "daily productivity driver," Spark is intended for quick prototyping and real-time collaboration, contrasting with the heavier, long-task orientation of the main Codex model. It is currently available in a research preview for ChatGPT Pro users within the Codex application. OpenAI positions Spark as the initial step toward a dual-mode Codex, capable of both instantaneous iteration and deep, extended reasoning tasks.

In a statement, Cerebras CTO and co-founder Sean Lie expressed enthusiasm for the collaboration's potential: "Partnering with OpenAI and developers to explore what fast inference enables—new interactions, new applications, a fundamentally different experience—is what's most exciting. This preview is just the start."

The launch follows a period of significant growth for Cerebras, which recently secured $1 billion in new funding at a $23 billion valuation and has signaled plans for an eventual public offering.

Source: TechCrunch