Content
summary Summary

OpenAI has released GPT-5 Codex, a new variant of GPT-5 built for use as an automated coding agent. The model is designed to handle real development tasks like refactoring, generating tests, fixing bugs, and reviewing code. It's part of the updated Codex system, which runs inside terminals, IDEs, web environments, and GitHub.

Ad

According to OpenAI, GPT-5 Codex can take part in short, interactive coding sessions as well as long-running, independent tasks. In testing, it worked continuously for more than seven hours, iterating on complex problems and running multiple test cycles. The model adjusts its compute time depending on task difficulty - using fewer tokens than GPT-5 on simple tasks and more on harder ones.

Stronger results than GPT-5 on benchmarks

On standard benchmarks, GPT-5 Codex outperforms GPT-5. In the SWE-bench Verified dataset, it hits 74.5 percent accuracy compared to GPT-5's 72.8 percent. For refactoring tasks, GPT-5 Codex scores 51.3 percent, while GPT-5 manages just 33.9 percent. These benchmarks include challenges from established open-source projects written in languages like Python, Go, and OCaml.

The model was also trained specifically for code reviews. It checks commits, verifies dependencies, runs tests, and compares intended changes against actual diffs. According to OpenAI, experienced developers rated the output higher than GPT-5, with GPT-5 Codex providing fewer irrelevant or incorrect comments than other models.

Ad
Ad

Integration into development workflows

Codex is accessible through the command line, IDE extensions (including VS Code), and a cloud-based environment. In the CLI, developers can attach screenshots or diagrams to explain design choices. The system can generate to-do lists and connect to external tools such as web search or MCP servers.

The cloud version adds extra automation. It detects common setup scripts, installs missing dependencies, and can even launch its own browser to check visual output. Screenshots of results can be attached directly to tasks or pull requests.

GPT-5 Codex is included with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans. Higher-tier plans allow multiple projects per week, and Business and Enterprise users can purchase additional usage beyond their standard allocation. An API integration is planned but not available yet.

Ad
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • OpenAI has released GPT-5 Codex, a specialized coding variant of GPT-5 designed to function as an automated development agent that can handle real programming tasks like refactoring, bug fixes, test generation, and code reviews across terminals, IDEs, and GitHub.
  • The model outperforms standard GPT-5 on coding benchmarks, achieving 74.5% accuracy on SWE-bench Verified compared to GPT-5's 72.8%, and scoring 51.3% on refactoring tasks versus GPT-5's 33.9%, while dynamically adjusting compute resources based on task complexity.
  • GPT-5 Codex integrates into development workflows through command line tools, IDE extensions, and cloud environments that can automatically detect setup scripts, install dependencies, and launch browsers for visual testing, with access included in ChatGPT Plus and higher-tier plans.
Sources
Max is the managing editor of THE DECODER, bringing his background in philosophy to explore questions of consciousness and whether machines truly think or just pretend to.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.