How to use GPT5 Codeex for autonomous coding?

TL;DR

GPT5 Codeex is a GPT5 variant optimized for coding tasks inside Codeex, enabling long autonomous sessions and improved code review. It can run in the terminal, IDEs, GitHub, and the ChatGPT iOS app, with faster task completion and reduced latency. It can automatically set up environments and iterate on code with tests.

Transcript

OpenAI just dropped GPT5 Codeex. If you thought Codeex was a great product, now it's powered by GPT5. It was only a matter of time. So, if you click the little get started button, look at this. VS Code Cursor, Windsor, VS Code Marketplace, Codeex Web. Today, we're releasing GPT5 Codeex, a version of GPT5, further optimized for agentic coding in Cod... Read More

Key Insights

• GPT5 Codeex is a version of GPT5 optimized for agentic coding in Codeex.
• Benchmarks show GPT5 Codeex scoring higher than GPT5 on SWEBench, with 74.5 versus 72.8.
• Codeex can work autonomously for long periods, with testing, debugging, and iteration continuing for hours.
• Codeex provides code reviews by reasoning about dependencies, running code, and validating behavior with tests.
• The tool reduces token usage on simpler turns while spending more resources on complex tasks.
• Codeex can operate in terminal, IDEs, GitHub, and the ChatGPT iOS app, expanding where you develop.
• Cloud infrastructure performance is improved with 90 percent lower latency and faster task completion.
• Windsurf is promoted as an IDE integration in the sponsor segment, enabling a smoother workflow.
• More videos from this channel:
  • How to run six open source AI tools locally today
  • Is Gemini 3.8 Flash the Best Low-Cost AI Model?
  • What Is Jev and How Does Its AI Model Work?
• More from this channel: Matthew Berman

Questions & Answers

Q: What is GPT5 Codeex and how is it different from GPT5 for coding tasks?

GPT5 Codeex is a version of GPT5 that is specifically trained and optimized for agentic coding in Codeex. It focuses on real world software engineering work, supports interactive sessions as well as long, complex tasks, and includes built in code review capabilities. This combination enables more automated coding workflows and faster iterations in development tasks.

Q: How does GPT5 Codeex perform in benchmarks compared to GPT5?

In the benchmarks mentioned, GPT5 Codeex scores higher than GPT5 on SWEBench, with a reported score of 74.5 versus 72.8 for GPT5. This indicates an improvement in performance metrics that relate to code-related tasks and efficiency, though the difference is described as a couple of points rather than a dramatic leap.

Q: What environments can Codeex operate in according to the video?

Codeex can operate wherever you develop, including in your terminal or IDE, on the web via GitHub, and even in the ChatGPT iOS app. This cross platform capability is highlighted as a feature that lets developers use Codeex across different parts of their workflow without being confined to a single interface.

Q: How long can GPT5 Codeex run autonomously and what does it do during that time?

GPT5 Codeex has been tested to run independently for more than 7 hours on large, complex tasks. During this time it iterates on implementations, fixes test failures, and delivers a successful implementation, performing coding, testing, debugging, and validation largely on its own.

Q: What makes GPT5 Codeex better at code reviews than typical tools?

GPT5 Codeex has been trained specifically for conducting code reviews by navigating codebases, reasoning about dependencies, running code and tests, and validating behavior. It can post analysis on pull requests and even be invoked to review for vulnerabilities, filling gaps left by human reviewers.

Q: What UI or workflow improvements are highlighted for Codeex.cli and integration?

The Codeex CLI has been updated with a terminal UI that includes tool calls and better formatted diffs. Approval modes are simplified into three levels, and the system supports compacting conversation state to manage longer sessions, making it easier to integrate into daily development workflows.

Q: How does Codeex handle environment setup and dependencies?

Codeex automatically sets up its own environment by scanning for common setup scripts and executing them. It can fetch dependencies at runtime using commands like pip install and can spin up its own browser to view results, attach screenshots to tasks, and integrate with GitHub PR workflows.

Q: What is the pricing and plan context for Codeex usage as described?

The video notes that Codeex is available for various plans including ChatGPT Plus, Pro, Business, Educational, and Enterprise. It mentions specific pricing for Pro and describes how different plans offer different levels of usage, including shared credits on enterprise plans and the ability to purchase credits for additional usage.

Summary & Key Takeaways

• GPT5 Codeex is a version of GPT5 optimized for agentic coding in Codeex. It powers autonomous coding sessions and code reviews with test validation.

• Benchmarks show GPT5 Codeex scoring higher than GPT5 on SWEBench, and it excels at refactoring, testing, and iterating across large tasks.

• Codeex can operate across multiple environments, automatically set up its workspace, and provide integrated PR reviews, improving developer velocity and code quality.