OPENAI
This is for the speed addicts
On Thursday, OpenAI released a lighter and faster version of its coding tool Codex, called GPT-5.3-Codex-Spark.
OpenAI says this model is built to respond more quickly, making it better suited for real-time coding help.
To achieve that speed, OpenAI is using a specialised chip from its hardware partner Cerebras, showing a closer connection between OpenAI’s AI models and the physical systems running them.
OpenAI and Cerebras announced their partnership last month through a multi-year deal reportedly worth over $10 billion.
OpenAI calls Spark the first major step in that agreement.
Codex-Spark runs on Cerebras’ Wafer Scale Engine 3, a chip with 4 trillion transistors.
OpenAI says Spark is meant for everyday coding tasks like quick prototyping and fast back-and-forth work, rather than longer and more complex jobs handled by the full Codex model.
Spark is currently available as a research preview for ChatGPT Pro users in the Codex app.
CEO Sam Altman hinted at the launch earlier, posting that something new was coming for Pro users and that it “sparks joy.”
In short:
OpenAI launched Codex-Spark, a faster coding model for quick workflows.
It uses Cerebras chips, marking a deeper partnership between the two companies.
The focus is on real-time AI coding support, not long-running tasks.
Speed is the whole point
OpenAI says Spark is part of a bigger plan for Codex to work in two modes: one for fast collaboration, and another for deeper reasoning and longer tasks.
Cerebras, which has become more important in the AI chip space, recently raised $1 billion at a valuation of $23 billion, and has said it may pursue an IPO in the future.
Cerebras CTO Sean Lie said the company is excited to see what developers can build with much faster AI responses, calling this preview only the beginning.
The real headline is: latency is the new flex.- MG