FastMCP FastMCP
Sign In
Back to Blog

GPT-5-Codex: The Revolution in Agentic Programming

• 2 views

OpenAI continues to amaze developers — the company has announced a major update to its AI coding agent, Codex. The new version is powered by GPT-5-Codex, a specialized version of GPT-5 optimized for agentic coding.

What's New in GPT-5-Codex

Dynamic execution time planning is the key feature of the new model. Unlike traditional routers that immediately determine required resources, GPT-5-Codex can adapt during execution. The model can start with a simple task taking seconds, then decide it needs to spend an additional hour or even seven hours on a deeper solution.

Expanded ecosystem usage — Codex now works everywhere: in the terminal, IDE, web interface, GitHub, and even in the ChatGPT iOS mobile app. This creates a seamless development experience where you can start a task in one place and continue in another.

Integration with Development Tools

The deep integration with existing developer workflows is particularly impressive:

  • VS Code extension for working directly in the editor
  • GitHub integration through simple @codex mentions in pull requests
  • Automatic code reviews with detailed change analysis
  • CLI interface for terminal-based work

The model supports internet access during task execution, allowing it to install dependencies, update packages, and run tests with external resources.

MCP Support in FastMCP

Important news for the MCP (Model Context Protocol) ecosystem — Codex CLI supports MCP servers. This opens new possibilities for integration with various tools and services.

And in FastMCP, this support is implemented as conveniently as possible — you can now install any MCP server with literally one click. This dramatically simplifies the setup and use of various tools in conjunction with Codex, allowing developers to quickly expand functionality without complex configuration file setup.

Performance and Code Quality

According to OpenAI's internal benchmarks, codex-1 shows strong performance even without AGENTS.md files or special configurations. The model was trained using reinforcement learning on real programming tasks, allowing it to generate code that closely mirrors human style.

Special attention was paid to creating "clean patches" ready for immediate review and integration into standard team development workflows.

Practical Improvements

Codex now caches containers to start new tasks 90% faster, reducing median startup time from 48 to 5 seconds. This significantly improves the user experience when working with the agent.

Added support for:

  • Binary files when applying patches
  • Voice input for tasks
  • Multiple responses for a single task
  • Canceling running tasks

Conclusions

GPT-5-Codex represents a serious step forward in AI-assisted development. Dynamic execution time planning combined with deep integration into existing tools makes it a powerful assistant for developers of any level.

MCP protocol support and simplified integration through FastMCP open new horizons for creating customized workflows. This isn't just a model update — it's a new paradigm for AI interaction in the development process.

All that remains is to wait for GPT-5-Codex availability through the API for integration into custom tools and services.