release

Introducing GPT-5.3-Codex-Spark

For builders of AI-powered development tools, this model demonstrates a push toward real-time assistance that can integrate into live coding environments, potentially reducing iteration cycles. It also sets a new performance benchmark for coding-specific LLMs.

OpenAI Blog·February 12, 2026·1 min readrelease

releaseIntroducing GPT-5.3-Codex-Spark

openai.com

What happened

OpenAI has released GPT-5.3-Codex-Spark, a new model designed for real-time code generation. According to OpenAI Blog, it offers 15x faster generation than previous versions and supports a 128k token context window. The model is currently available as a research preview exclusively for ChatGPT Pro subscribers. This release marks a significant step in optimizing AI for interactive coding tasks, where speed is critical. For developers and solopreneurs building AI workflows, the key takeaway is the potential for near-instantaneous code suggestions in real-time editors, reducing latency in iterative development. While the model is proprietary and limited to ChatGPT Pro, it signals a trend toward specialized, high-speed coding assistants. The context window expansion also allows handling larger codebases in a single pass, which could improve contextual understanding in complex projects. However, access constraints mean widespread adoption hinges on broader availability.

Key takeaways

OpenAI introduced GPT-5.3-Codex-Spark, a real-time coding model with 15x faster generation than prior versions.
The model features a 128k token context window, enabling it to process larger code snippets or files at once.
It is currently in research preview and available only to ChatGPT Pro subscribers.
The model is optimized for interactive, low-latency code generation tasks.
This release builds on OpenAI's Codex series, focusing on speed and context length for coding workflows.