Key Highlights:
- Focuses on low-latency, real-time coding workflows for Pro users.
- The new model runs on Cerebras’ dedicated Wafer Scale Engine 3 chip.
- This marks the first major milestone in OpenAI’s multi-year, $10 billion-plus partnership with Cerebras.
OpenAI has launched a new lightweight version of Codex, powered by a dedicated Cerebras chip, to deliver faster AI-driven coding assistance. The update matters because it shows how OpenAI is reshaping its infrastructure to cut latency and support real-time developer workflows, not just larger, slower AI tasks.
OpenAI rolls out faster Codex with custom hardware support
OpenAI announced GPT-5.3-Codex-Spark on Thursday. The model is a smaller, faster variant of its latest Codex system, released earlier this month. Unlike the standard version, Codex-Spark is optimized for speed and responsiveness.
To achieve that, OpenAI is using a dedicated chip from its hardware partner, Cerebras. This marks a deeper level of hardware-software integration for OpenAI as it looks to scale AI tools for everyday developer use.
What is GPT-5.3-Codex-Spark?
Codex-Spark is designed for rapid iteration. OpenAI describes it as a daily productivity tool rather than a model meant for long-running, complex coding jobs.
The model focuses on:
- Fast inference
- Real-time collaboration
- Rapid prototyping and feedback loops
As a result, it targets developers who want instant responses while coding, debugging, or testing ideas.
Why Cerebras’ chip matters here
Codex-Spark runs on Cerebras’ Wafer Scale Engine 3, or Wafer Scale Engine 3. The chip packs 4 trillion transistors and is built specifically for AI workloads that demand extremely low latency.
OpenAI says this hardware choice helps Codex-Spark respond faster than models running on traditional GPU-based infrastructure. Therefore, it supports workflows where speed directly affects productivity.
A milestone in OpenAI–Cerebras partnership
The Cerebras partnership was announced last month as a multi-year deal valued at over $10 billion. At the time, OpenAI said the goal was simple: make AI responses much faster.
Now, OpenAI is calling Codex-Spark the first milestone in that relationship. It also signals a shift toward using specialized chips alongside conventional compute for different AI tasks.
Who gets access?
Codex-Spark is currently available as a research preview for ChatGPT Pro users through the Codex app.
Ahead of the launch, OpenAI CEO Sam Altman hinted at the release on social media, calling it “a special thing” that “sparks joy.”
Why this Codex update is important
OpenAI says Codex-Spark is the first step toward a dual-mode Codex. One mode supports real-time collaboration, while the other handles deeper, long-running tasks. By pairing Codex with dedicated hardware, OpenAI is signaling that speed, not just scale, will define the next phase of AI coding tools.
In short, Codex is no longer just about writing code. It is about responding fast enough to work alongside developers, in real time.
