OpenAI Launches GPT-5.3-Codex-Spark on Cerebras Hardware

OpenAI launched GPT-5.3-Codex-Spark on February 12, 2026, as a compact AI model optimized for real-time software development. The model represents OpenAI's first major production deployment on hardware from Cerebras Systems Inc., utilizing Wafer Scale Engine 3 accelerators to deliver over 1,000 tokens per second. This speed is up to 15 times faster than traditional AI coding models, enabling near-instant conversational assistance for targeted edits and logic refinement.

To support this low-latency experience, OpenAI redesigned its infrastructure with persistent WebSocket connections and an optimized Responses API, which reduced client-server roundtrip overhead by 80%. While the Spark model offers a 128,000-token context window, it is a slimmed-down version of the general-purpose coding software and possesses lower fidelity and fewer capabilities for complex autonomous tasks.

This deployment follows a deal exceeding US$10 billion and is part of a broader strategic effort to diversify OpenAI's hardware ecosystem and reduce its dependence on Nvidia Corp. Although Nvidia remains the core of the company's training and inference stack, OpenAI is expanding its partnerships to include Advanced Micro Devices Inc. and Broadcom Inc. The model is currently available as a research preview for ChatGPT Pro subscribers via the Codex app, CLI, and VS Code extension, with enterprise API access rolling out to select partners.