OpenAI says GPT-5.3-Codex-Spark is its first AI model that runs on Cerebras chips, after they signed a $10B+ deal in January; Codex has 1M+ weekly active users

at 1,000 tokens/s. [video]@openaidevs:Introducing GPT-5.3-Codex-Spark, our ultra-fast model purpose built for real-time coding. We're rolling it out as a research preview for ChatGPT Pro users in the Codex app, Codex CLI, and IDE extension. [video]Ben Bajarin /@benbajarin:As the world moves to inference, dedicated inference designs will be prominant. Great customer case for @cerebras

Bloomberg 2026-02-13 Rachel Metz

Context & Ripple Effects

OpenAI had just positioned GPT-5.3-Codex as a faster coding model for longer-running tasks; that earlier Codex release established the product base that Spark now extends into real-time use.

The Spark preview puts the new capability in the Codex app, CLI and IDE extension, bringing a hardware integration to a coding product OpenAI says already has more than 1 million weekly active users.

First-order effects

OpenAI gains a Cerebras-backed serving path for GPT-5.3-Codex-Spark, initially exposing its high-speed coding experience to ChatGPT Pro users across its developer tools.
Cerebras gets a live OpenAI model deployment tied to a widely used coding product, rather than only a commercial infrastructure agreement.

Second-order effects

For coding-model providers, responsiveness becomes a more visible product dimension alongside code quality and task completion, especially in interactive IDE and CLI workflows.
OpenAI can assess whether specialized inference hardware improves the user experience and operating profile of a production coding service before broadening access.

Third-order effects

If deployments like this expand, leading AI services may increasingly use heterogeneous compute—matching models or workloads to different hardware rather than relying on a single chip supplier.
That would shift AI infrastructure competition toward demonstrable inference performance in end-user products, not just model-training capacity.

The trend: AI providers are turning specialized inference hardware into a product differentiator for latency-sensitive developer tools.

Discussion

@kylebrussell Kyle Russell on x
I thought this was going to come like next year, not now
@openaidevs @openaidevs on x
GPT-5.3-Codex-Spark is the first milestone in our partnership with @cerebras. It provides a faster tier on the same production stack as our other models, complementing GPUs for workloads where low latency is critical. https://openai.com/...
@mweinbach Max Weinbach on x
Codex Spark was trained on GPUs for Cerebras hardware but OpenAI added support to their inference framework for Cerebras meaning they're reading to load future models onto it too GPUs are still foundational for inference and training, though [image]
@cerebras @cerebras on x
OpenAI Codex-Spark powered by Cerebras You can now just build things faster—at 1,000 tokens/s. [video]
@openaidevs @openaidevs on x
Introducing GPT-5.3-Codex-Spark, our ultra-fast model purpose built for real-time coding. We're rolling it out as a research preview for ChatGPT Pro users in the Codex app, Codex CLI, and IDE extension. [video]
@benbajarin Ben Bajarin on x
As the world moves to inference, dedicated inference designs will be prominant. Great customer case for @cerebras

Chronicles