OpenAI launches GPT-5.3-Codex-Spark powered by Cerebras hardware

OpenAI just shipped a product that doesn’t run on Nvidia hardware. That sentence alone is worth pausing on.

GPT-5.3-Codex-Spark, a smaller and faster variant of OpenAI’s coding model, is now live for ChatGPT Pro users. It runs exclusively on Cerebras Systems’ Wafer Scale Engine 3, generating over 1,000 tokens per second with a 15x speed improvement over the standard GPT-5.3-Codex model.

What Codex-Spark actually does

The “Spark” designation reflects the design philosophy. This isn’t a model meant to architect entire codebases from scratch. It’s optimized for real-time collaboration, the kind where a developer writes a line and gets an intelligent suggestion before their fingers leave the keyboard. Users can even interrupt the model mid-generation, something that matters when you’re iterating at speed and don’t want to wait for a response you’ve already moved past.

Access comes through the Codex desktop app, command-line interface, and VS Code extension. Each channel has separate usage meters, suggesting OpenAI expects developers to use Spark differently depending on their workflow.

For now, the tool is limited to ChatGPT Pro subscribers. OpenAI hasn’t indicated when, or whether, it will trickle down to lower tiers.

The Cerebras angle matters more than the model itself

OpenAI has been, for all practical purposes, an Nvidia shop. Codex-Spark is the first OpenAI product to break that pattern entirely, running on Cerebras’ WSE-3 chips instead.

Cerebras takes a fundamentally different approach to AI compute. Where Nvidia builds discrete GPUs that get networked together in massive clusters, Cerebras manufactures wafer-scale processors. A single WSE-3 chip is roughly the size of a dinner plate, containing hundreds of thousands of cores on one piece of silicon. The architecture eliminates much of the communication overhead that comes from linking thousands of separate GPUs, which is partly why the latency numbers for Spark are so aggressive.

The 15x acceleration over standard Codex isn’t just a software optimization. It’s a hardware story.

What this means for crypto and AI-adjacent markets

Neither OpenAI nor Cerebras has a token. Neither company is publicly traded as of this writing. So the direct market impact on crypto portfolios is zero.

Real-time coding assistance at this speed has obvious applications for blockchain development. Writing smart contracts is painstaking work where bugs can be catastrophic, sometimes to the tune of hundreds of millions of dollars in exploited protocols. A tool that can suggest, review, and iterate on Solidity or Rust code in real time could meaningfully reduce the error rate in DeFi development.

OpenAI has signaled that future versions of Codex will integrate both real-time capabilities like Spark and long-horizon planning from the full Codex model.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.