OpenAI has introduced GPT-5.3-Codex-Spark, the research preview of our ultra-fast coding model optimized for rapid software development, said OpenAI, a tremendous milestone in the development of on-demand AI coding assistance. GPT-5.3-Codex-Spark is the streamlined version of the GPT-5.3-Codex that operates on the ultra-low-latency inference of Cerebras Wafer-Scale Engine 3 hardware, boasting an ultra-fast velocity of more than 1,000 tokens per second and offering “interruptible” capability to modify the AI’s flow of execution to address targeted code changes, logic restructurings, interface improvements, and other processes to speedily respond to the needs of the coder or programmer with rapid feedback on code or edits made while the AI is “executing.”
Also Read: Kyndryl Launches Agentic AI Workflow Governance for Trusted Deployment
At launch, the model features a 128k token context window and will be available to ChatGPT Pro users in the Codex app, CLI and IDE extensions, with separate rate limits during the preview phase as OpenAI scales capacity and hardens the user experience. OpenAI’s engineers and partners will collect developer feedback to guide future enhancements, including support for larger models, multimodal inputs, and broader integration across tools and workflows.