Introducing Upgrades to Codex | OpenAI

Summary: OpenAI unveils GPT-5-Codex, a specialized version of GPT-5 for advanced coding tasks, integrated into Codex's ecosystem. This update enhances code review, autonomous task execution, and developer collaboration across CLI, IDEs, and GitHub. Safety features, pricing plans, and performance optimizations are also highlighted, positioning Codex as a reliable coding partner.

---

1. The Launch of GPT-5-Codex: A Leap in Agentic Coding

Today, OpenAI announced GPT-5-Codex, a version of GPT-5 fine-tuned for agentic software engineering. Unlike its general-purpose sibling, this model excels at real-world coding tasks, from building projects from scratch to debugging and large-scale refactors.

"GPT-5-Codex is equally proficient at quick, interactive sessions and independently powering through long, complex tasks."

Key features include:

• Dynamic thinking: Adjusts time spent on tasks based on complexity. For example, it worked autonomously for 7+ hours on large refactors during testing.
• Efficient token usage: Uses 93.7% fewer tokens for simple tasks but doubles reasoning time for complex ones.
• Code review mastery: Trained to catch critical bugs by analyzing dependencies and running tests. OpenAI's internal testing showed it reduces unimportant or incorrect review comments, letting developers focus on high-impact issues.

For frontend tasks, GPT-5-Codex now generates mobile websites with higher human preference scores and can analyze images or screenshots in the cloud.

---

2. Codex Updates: A Seamless Developer Experience

Since its launch in 2024, Codex has evolved into a unified tool accessible via ChatGPT, CLI, IDE extensions, and GitHub. Recent updates include:

Revamped Codex CLI and IDE Extension

• Image integration: Share screenshots, wireframes, or diagrams directly in the CLI to build design context.
• Progress tracking: Automatically generates to-do lists for complex tasks.
• Tool enhancements: Web search, MCP (Multi-Tool Protocol), and simplified approval modes (read-only, auto, full access).
• Cloud-IDE sync: Start a task in the cloud and continue editing it locally without losing context.

"When working in the cloud, it can look at images or screenshots you provide as input, visually inspect its progress, and display screenshots of its work to you."

The IDE extension (for VS Code, Cursor, etc.) offers faster results by leveraging open files and selected code, reducing the need for lengthy prompts.

Cloud Performance Boosts

• 90% faster completion times via container caching.
• Automatic environment setup: Detects and runs setup scripts, installs dependencies (e.g., pip install).
• Image-based feedback: Spins up browsers, takes screenshots, and attaches them to tasks or GitHub PRs.

---

3. Code Review Revolution: Precision and Context Awareness

GPT-5-Codex's code review capabilities outperform static analysis tools by:

1. Matching PR intent to actual code changes.
2. Reasoning over the entire codebase and dependencies.
3. Executing tests to validate behavior.

At OpenAI, Codex now reviews most PRs, catching hundreds of issues daily—often before human reviews. Developers can trigger reviews via @codex review in GitHub with custom guidance (e.g., "review for security vulnerabilities").

"Only the most thorough human reviewers put this level of effort into every PR they review, so Codex fills the gap—helping teams find problems earlier, reduce reviewer load, and ship with more confidence."

---

4. Safety and Security: Building Trustworthy AI

Codex prioritizes security with:

• Sandboxed environments: Network access is disabled by default, preventing harmful actions.
• Permission prompts: Asks for approval before risky operations (e.g., running commands).
• Customizable controls: Cloud users can restrict network access to trusted domains; CLI/IDE users can approve commands case-by-case.

OpenAI classifies GPT-5-Codex as High capability in biological/chemical domains and has implemented safeguards. Detailed safety measures are in the system card addendum.

"We always encourage developers to review the agent's work before making changes or deploying to production."

---

5. Pricing and Availability

Codex is included in ChatGPT Plus, Pro, Business, Edu, and Enterprise plans. Usage scales with your plan:

• Plus/Edu/Business: Suitable for weekly coding sessions.
• Pro: Supports a full workweek across projects.
• Enterprise: Shared credit pool for flexible scaling.

API access for Codex CLI via GPT-5-Codex is coming soon.

---

Conclusion: Codex as Your Ultimate Coding Teammate

With GPT-5-Codex, OpenAI has created a coding partner that's faster, smarter, and deeply integrated into developers' workflows. From autonomous refactors to AI-driven code reviews, Codex reduces errors, saves time, and adapts to both simple and complex tasks. As OpenAI continues to refine its safety and performance, Codex is poised to become an indispensable tool for ambitious projects—one that works with developers, not just for them.

🚀 Ready to build the future? Dive into Codex and see what your new AI teammate can do!