A new version of OpenAI’s Codex is powered by a new dedicated chip

On Thursday, OpenAI announced the merchandise of a light-weight mentation of its agentic coding instrumentality Codex, the latest exemplary of which OpenAI launched earlier this month. GPT-5.3-Codex-Spark is described by the institution arsenic a “smaller version” of that model, 1 that is designed for faster inference. To powerfulness that inference, OpenAI has brought successful a dedicated spot from its hardware spouse Cerebras, marking a caller level of integration successful the company’s carnal infrastructure.

The partnership betwixt Cerebras and OpenAI was announced past month, erstwhile OpenAI said that it had reached a multi-year statement with the steadfast worthy implicit $10 billion. “Integrating Cerebras into our premix of compute solutions is each astir making our AI respond overmuch faster,” the institution said astatine the time. Now, OpenAI calls Spark the “first milestone” successful that relationship.

Spark, which OpenAI says is designed for swift, real-time collaboration and “rapid iteration,” volition beryllium powered by Cerebras’ Wafer Scale Engine 3. The WSE-3 is Cerebras’ third-generation waferscale megachip, decked retired with 4 trillion transistors. OpenAI describes the caller lightweight instrumentality arsenic a “daily productivity driver, helping users with accelerated prototyping” alternatively than the longer, heavier tasks that the archetypal 5.3 is designed for. Spark is presently enjoying a probe preview for ChatGPT Pro users successful the Codex app.

In a tweet successful beforehand of the announcement, CEO Sam Altman seemed to hint astatine the caller model. “We person a peculiar happening launching to Codex users connected the Pro program aboriginal today,” Altman tweeted. “It sparks joyousness for me.”

In its authoritative statement, OpenAI emphasized Spark arsenic designed for the lowest imaginable latency connected Codex. “Codex-Spark is the archetypal measurement toward a Codex that works successful 2 complementary modes: real-time collaboration when you privation accelerated iteration, and long-running tasks when you request deeper reasoning and execution,” OpenAI shared. The institution added that Cerebras’ chips excel astatine assisting “workflows that request highly debased latency.”

Cerebras has been astir for implicit a decade but, successful the AI era, it has enjoyed an progressively salient relation successful the tech industry. Just past week, the institution announced that it had raised $1 cardinal successful caller superior astatine a valuation of $23 billion. The institution has previously announced its intentions to prosecute an IPO.

“What excites america astir astir GPT-5.3-Codex-Spark is partnering with OpenAI and the developer assemblage to observe what accelerated inference makes imaginable — caller enactment patterns, caller usage cases, and a fundamentally antithetic exemplary experience,” Sean Lie, CTO and Co-Founder of Cerebras, said successful a statement. “This preview is conscionable the beginning.”