OpenAI has launched GPT-5.3-Codex-Spark, a new AI model focused on coding. It is built for real-time development and can generate code at speeds over 1,000 tokens per second.
The model is now available as a research preview. It is OpenAI’s first system made for low-latency coding, aiming to give developers almost instant responses in live settings.
Built for real-time coding performance
Interesting Engineering reported that the model’s main feature is its ability to produce code very quickly, which allows for real-time interaction during programming. Developers can see results almost right away as they write or change instructions.
Gadgets 360 reported that GPT-5.3-Codex-Spark is the first Codex model made mainly for instant code generation and editing. It is built for low-latency tasks and can write, refine, and adjust code almost instantly, while keeping a context window of about 128,000 tokens.
The outlet also said the model is now available to ChatGPT Pro users through the Codex app, command-line tools, and IDE extensions, but it is still in a limited research preview.
Hardware shift enables higher speed
Business Today reports that the model’s speed comes in part from using special low-latency hardware. This lets it process over 1,000 tokens per second, which is much faster than earlier coding models.
GPT-5.3-Codex-Spark uses Cerebras’ Wafer Scale Engine 3 hardware instead of traditional GPUs for this kind of work. This setup is meant to reduce response times for interactive tasks, giving developers faster feedback.
The system is positioned as a faster, lighter alternative to the larger GPT-5.3-Codex model, prioritizing responsiveness and iteration speed over deep reasoning on complex software projects.
Focus on rapid iteration over complex reasoning
Gadgets 360 reported that Codex-Spark works well for quick edits, logic changes, and interface updates, but more complex tasks may still need the full GPT-5.3-Codex model. The company’s internal tests show it performs well on software engineering tests and keeps faster output speeds.
The model was built to support developers who need rapid, incremental changes during active coding sessions, allowing them to interrupt workflows and get immediate updates.
Expansion of OpenAI’s coding ecosystem
The release comes as OpenAI increases its focus on coding-specific AI systems. Gadgets 360 reported that earlier in 2026, company leadership highlighted rapid growth in usage of Codex tools, and the new model extends that push by emphasizing speed and real-time interaction.
Interesting Engineering reported that the new system shows a move toward AI tools that not only generate code, but also act as ongoing partners in live development.
Share
Copied
