OpenAI is shaking things up with its latest release, GPT-5.3 Codex-Spark, which runs on Cerebras chips instead of the usual Nvidia ones. This is no small shift. The model churns out code at an impressive rate of over 1,000 tokens per second. That's around 15 times faster than its predecessor.
Breaking the Speed Barrier
Here's what the benchmarks actually show: Codex-Spark's speed stands in stark contrast to Anthropic's Claude Opus 4.6. Even in its premium fast mode, Claude Opus only manages to reach speeds 2.5 times its standard 68.2 tokens per second. Clearly, OpenAI is setting a new standard for speed in AI coding models.
Sachin Katti, OpenAI's head of compute, expressed enthusiasm over the collaboration with Cerebras. "Cerebras has been a great engineering partner," he noted, highlighting the addition of fast inference as a important platform capability. Frankly, this partnership marks a significant milestone in AI hardware diversification.
What's in the Package?
Codex-Spark, currently in research preview, is available to ChatGPT Pro subscribers for $200 a month. Users can access it via the Codex app, command-line interface, and VS Code extension. With a substantial 128,000-token context window, it handles text-only tasks at launch. OpenAI is also rolling out API access to select partners, which could broaden its use cases.
Why This Matters
Strip away the marketing and you get a clear picture: this move signifies a bold step away from Nvidia's monopoly in AI hardware. But who benefits the most here? Certainly, developers looking for faster coding solutions. With AI models becoming increasingly integral to software development, the speed and efficiency of Codex-Spark can be a big deal.
Yet, one can ask, will this shift in hardware preference influence other AI developers to explore non-Nvidia options? The reality is, if OpenAI's gamble pays off, we might see a broader adoption of alternative chip manufacturers like Cerebras.
OpenAI's decision is more than just a technical feat. It's a strategic pivot that could reshape the AI hardware market. The architecture matters more than the parameter count, and Codex-Spark is proof of that.
