GPT-5.3-Codex-Spark Unveiled

OpenAI has released a research preview of GPT-5.3-Codex-Spark, a small model optimized for real-time coding. Developed through a partnership with Cerebras, this model generates over 1,000 tokens per second, delivering near-real-time response speed.

Unlike existing models that perform large-scale tasks, Codex-Spark specializes in immediate interactions such as real-time code editing and logic restructuring. It supports a 128k context window and has demonstrated strong performance on the SWE-Bench Pro and Terminal-Bench 2.0 benchmarks.

In addition, the following technical improvements were introduced to reduce latency across the entire request-response pipeline:

• Introduction of WebSocket connections and optimization of the Responses API
• 80% reduction in client/server round-trip overhead
• 30% reduction in per-token overhead and 50% reduction in time to first token (TTFT)

This model runs on Cerebras's Wafer Scale Engine 3 accelerator. While GPUs are cost-effective for broad usage, Cerebras complements this by handling workflows that require extremely low latency, achieving optimal performance.

It is currently available as a research preview for ChatGPT Pro users in the Codex app, CLI, and VS Code extension.