GPT-5.3-Codex System Card

OpenAI has introduced GPT-5.3-Codex, an agentic coding model designed for long-running tasks involving research, tool use, and complex execution. The model combines the frontier coding performance of GPT-5.2-Codex with the reasoning and professional knowledge capabilities of GPT-5.2, allowing users to interact with and steer the model during active workflows without loss of context.

Agentic Coding Capabilities

GPT-5.3-Codex functions as an agentic tool capable of executing complex, multi-step technical projects. By merging the specialized coding strengths of the 5.2-Codex series with the general reasoning capabilities of GPT-5.2, the model can manage tasks that require sustained execution and the integration of professional knowledge. This architecture allows for a collaborative interaction model where the user can provide guidance while the model performs research and utilizes tools to complete a task.

Safety and Preparedness Framework

OpenAI has applied specific safety classifications and safeguards to GPT-5.3-Codex based on its internal Preparedness Framework:

Biology and AI Self-Improvement

GPT-5.3-Codex is classified as having "High capability" in biology and is deployed with the standard suite of safeguards used across the GPT-5 family. However, the model does not reach the "High capability" threshold for AI self-improvement.

Cybersecurity Safeguards

GPT-5.3-Codex is the first launch that OpenAI is treating as "High capability" in the cybersecurity domain. While OpenAI states there is no definitive evidence that the model has reached the high threshold, the company is taking a precautionary approach. To mitigate risks, a layered safety stack has been implemented to impede and disrupt potential threat actors, with the goal of ensuring these capabilities remain accessible to cyber defenders.