OpenAI Unveils Real Time Coding Model Codex Spark

OpenAI has intensified its focus on coding-centric artificial intelligence in 2026 with the introduction of GPT 5.3 Codex Spark, its first model capable of writing and editing code in real time. The launch follows the earlier release of GPT 5.3 Codex, signalling a strategic shift toward developer-first tools. CEO Sam Altman recently described Codex’s rapid adoption as “insane,” noting a 50 percent surge in usage within a week.

Currently available in research preview for ChatGPT Pro users, Codex Spark can generate, modify, and refine code almost instantly, processing up to 1,000 tokens per second. Designed for low  latency workloads, the model features a 128,000 token context window and supports real-time logic adjustments, interface refinements, and targeted edits. While highly responsive, it is positioned for routine to moderately complex tasks, with heavier workloads still better suited to the flagship Codex model.

To achieve its speed, the model runs on specialised infrastructure developed in partnership with Cerebras, leveraging the Wafer Scale Engine 3 accelerator for high-speed inference. Internal benchmarks show Codex Spark outperforming earlier lightweight models while approaching the performance of GPT 5.3 Codex. OpenAI plans to expand access gradually and is also offering the model to select design partners via API to evaluate real-world developer integration.