OpenAI has introduced GPT-5.3-Codex-Spark, a lightweight version of its latest Codex model — and this time, it’s backed by a dedicated AI chip designed to dramatically speed up performance.

The move marks a deeper integration between OpenAI and its hardware partner Cerebras, signaling that AI software innovation is now tightly intertwined with physical compute infrastructure.

What Is GPT-5.3-Codex-Spark?

According to OpenAI, GPT-5.3-Codex-Spark is a smaller, faster variant of the recently released GPT-5.3 Codex model. The focus is clear: low latency and rapid iteration.

While the full 5.3 version is built for heavier, long-running tasks requiring deeper reasoning and execution, GPT-5.3-Codex-Spark is positioned as a daily productivity engine. It’s designed to help developers prototype quickly and collaborate in real time.

OpenAI described the new release as the “first milestone” in its expanded partnership with Cerebras.

Powered by Cerebras’ Wafer-Scale Engine 3

What sets GPT-5.3-Codex-Spark apart is its hardware foundation.

The model runs on Cerebras’ Wafer Scale Engine 3 (WSE-3), a third-generation megachip containing 4 trillion transistors. The chip is specifically optimized for ultra-low latency inference — the kind of performance developers need for interactive coding sessions.

Last month, OpenAI announced a multi-year agreement with Cerebras valued at over $10 billion. At the time, OpenAI emphasized that integrating Cerebras into its compute mix was aimed at significantly reducing response times.

GPT-5.3-Codex-Spark now represents the first tangible outcome of that agreement.

Built for Real-Time Collaboration

OpenAI says GPT-5.3-Codex-Spark is intended for “swift, real-time collaboration” and “rapid iteration.” In practice, that means faster feedback loops for developers working inside the Codex app.

The company framed the launch as part of a broader vision:

Codex will operate in two complementary modes — real-time collaboration for rapid iteration and long-running execution for deeper reasoning tasks.

For ChatGPT Pro users, GPT-5.3-Codex-Spark is currently available as a research preview inside the Codex app.

A Hint From Sam Altman

Ahead of the official announcement, OpenAI CEO Sam Altman teased the launch on social media, writing that something special was coming to Codex Pro users and adding, “It sparks joy for me.”

The name “Spark” now makes sense.

Why Low Latency Matters

In coding workflows, speed isn’t just about convenience — it changes how developers think and work.

Ultra-fast inference enables:

• Rapid prototyping
• Instant feedback on code changes
• Real-time collaborative editing
• Reduced cognitive friction

Cerebras’ architecture is designed to excel in precisely these low-latency environments, making GPT-5.3-Codex-Spark a hardware-optimized model rather than just a scaled-down variant.

Cerebras’ Rising Profile

Cerebras has existed for more than a decade, but its prominence has surged alongside the AI boom.

Just last week, the company raised $1 billion in new capital at a $23 billion valuation. It has also previously expressed interest in pursuing an IPO.

Cerebras CTO and co-founder Sean Lie highlighted the broader implications of GPT-5.3-Codex-Spark, noting that fast inference could unlock new interaction patterns and use cases that weren’t previously feasible.

“This preview is just the beginning,” he said.

A New Era of AI-Hardware Integration

The release of GPT-5.3-Codex-Spark underscores a growing reality in the AI race: breakthroughs are no longer driven solely by model architecture.

They’re increasingly dependent on specialized hardware.

With GPT-5.3-Codex-Spark, OpenAI isn’t just shipping a faster coding assistant — it’s demonstrating how deeply AI models and dedicated chips are becoming intertwined.

And for developers, that could mean a fundamentally different experience.