GPT-5.3 Codex

by OpenAI

OpenAI’s most capable agentic coding model with 25% faster inference and interactive real-time steering

See https://openai.com/blog/gpt-53-codex

Features

  • 25% faster inference than GPT-5.2-Codex
  • Interactive agentic coding with real-time steering and supervision
  • Multi-language software engineering (surpasses SWE-Bench Pro)
  • Robust terminal and OS interaction (Terminal-Bench 2.0, OSWorld-Verified)
  • Production-quality web and frontend generation
  • Advanced vulnerability detection and cybersecurity analysis
  • Computer use and desktop automation capabilities
  • Multi-file codebase navigation and manipulation

Superpowers

GPT-5.3-Codex is the ideal tool for teams building autonomous coding agents that need real-time human oversight. Unlike previous models that work invisibly, Codex provides frequent updates on key decisions, allowing users to steer complex tasks without losing context. The 25% speed improvement makes it practical for interactive workflows where responsiveness matters.

Best for:

  • Autonomous coding agents with human-in-the-loop control
  • Full-stack web development (frontend and backend)
  • System administration and DevOps automation
  • Security research and vulnerability analysis
  • Multi-file refactoring and codebase modernization
  • Data analysis pipeline creation

Security-First: First OpenAI model classified as “High capability” for cybersecurity—comes with enhanced access controls and restricted API deployment through “Trusted Access for Cyber” pilot program.

Availability

  • Immediate: All paid ChatGPT users via Codex app, CLI, IDE extensions, web
  • Rolling: Full API access with security restrictions for high-risk cybersecurity uses
  • Enable steering: Settings → General → Follow-up behavior in Codex app

Pricing

Pricing details not yet published. Full API access coming soon with security safeguards for cybersecurity use cases.