OpenAI introduces GPT‑5.3‑Codex‑Spark for real-time coding with 128k context window

OpenAI has announced GPT‑5.3‑Codex‑Spark, a compact model designed for real-time coding tasks. The research preview is now available and is the first in the Codex lineup created specifically to enable instant, interactive coding experiences.

The standout feature of Codex-Spark is its ultra-low latency, consistently delivering more than 1000 tokens per second on supported hardware. This allows for a highly responsive workflow, as users can perform targeted edits, adjust logic, or refine interfaces and see the effects immediately. These improvements mean that Codex now accommodates both extended, ambitious development tasks and short, in-the-moment interactions.

At launch, Codex-Spark offers a 128k-token context window and is released as a text-only model. Early access is available to ChatGPT Pro subscribers through the latest versions of the Codex app, command-line interface, and Visual Studio Code extension. During the research preview, Codex-Spark operates under its own rate limits, so usage does not count against standard plan quotas. However, access may be temporarily limited during periods of high demand. This rollout focuses on supporting highly interactive work, allowing users to collaborate in real time, redirect the model on the fly, and benefit from near-instant feedback throughout the coding process.