OpenAI has just taken another big swing at reshaping how developers interact with artificial intelligence. On February 12, the company unveiled GPT‑5.3‑Codex‑Spark, a leaner, ultra‑fast sibling of its flagship GPT-5.3-Codex model. The pitch is simple but ambitious: make coding with AI feel less like waiting for a response and more like collaborating in real time.

Codex‑Spark is built for speed. Running on Cerebras’ Wafer Scale Engine 3—a massive AI accelerator designed for low‑latency inference—the model can churn out more than 1,000 tokens per second. That’s not just a technical brag; it means developers can tweak a function, reshape logic, or refine an interface and see results almost instantly. OpenAI describes this as the first milestone in its multi‑year partnership with Cerebras, a deal reportedly worth over $10 billion.

The model itself is smaller than GPT-5.3-Codex, but it’s tuned to excel at interactive tasks where responsiveness matters as much as raw intelligence. Think of it as the difference between a marathon runner and a sprinter: while larger models are designed to autonomously tackle long‑running projects over hours or days, Codex‑Spark thrives in short bursts of rapid iteration. It doesn’t automatically run tests unless asked, and it keeps edits lightweight—perfect for developers who want to stay in the loop and guide the process step by step.

OpenAI has also reworked the plumbing behind the scenes. By streamlining its inference stack and introducing persistent WebSocket connections, the company claims to have cut client‑server round-trip overhead by 80%, per‑token overhead by 30%, and time‑to‑first‑token by 50%. In practice, this means Codex‑Spark doesn’t just think faster—it feels faster, tightening the feedback loop between human and machine.

For now, Codex‑Spark is text‑only with a 128k context window, and it’s rolling out as a research preview to ChatGPT Pro users through the Codex app, CLI, and VS Code extension. Select design partners are also getting early API access to explore how the model might slot into their products. OpenAI is clear that this is just the beginning: future iterations will expand into larger models, longer context lengths, and multimodal input.

The bigger picture here is about interaction speed. As models grow more capable, the bottleneck shifts from intelligence to responsiveness. Codex‑Spark is OpenAI’s attempt to break that bottleneck, making AI coding assistants feel more natural, more immediate, and ultimately more useful. It’s a move that could redefine how developers think about AI—not as a distant agent working in the background, but as a partner sitting right beside them, ready to iterate at the speed of thought.