GPT‑5-Codex Lands in Codex Tools, API Coming Soon, With Big Push on Code Review

OpenAI unveiled GPT‑5-Codex for coding tasks, integrated into VS Code, Codex CLI, and Codex Cloud, with API access coming later. Codex Cloud adds automated GitHub code reviews leveraging temporary containers. Early signals show improved refactoring and fewer low‑value comments, but some tool-usage gaps and rough edges remain.

Key Points

• GPT‑5-Codex is live in OpenAI’s coding tools (VS Code extension, Codex CLI, Codex Cloud) but not yet in the API.
• Codex Cloud now offers automated GitHub code reviews using ephemeral containers.
• Model upgrades focus on code review, dynamic compute time, refactoring quality, mobile-friendly output, and more focused comments.
• The GPT‑5-Codex system prompt is shorter than prior models’ prompts in Codex CLI.
• Early testing and reviews show promise but also shortcomings, such as poor use of the CLI’s search tool and imperfect creative outputs.

Sentiment

Mildly skeptical. While there is constructive technical discussion, the dominant thread involves questions about the credibility of those covering GPT-5, with accusations of shilling and concerns about the preview process. Simon Willison's transparent participation provides some balance but doesn't fully dispel the community's skepticism toward both the product and its coverage.

In Agreement

• Simon Willison confirms GPT-5 feels as impressive to him today as during the preview, particularly for code and search tasks
• The private demo genuinely seemed better than the public release, with many possible explanations beyond deception
• Code review improvements represent tangible capability gains

Opposed

• GPT-5-Codex may not truly be GPT-5 — could be a smaller or quantized model, and OpenAI's language is deliberately vague
• Tech influencers who attended OpenAI's private launch event served as marketing vehicles regardless of editorial independence claims
• OpenAI's preview process lacked transparency, with model IDs changing multiple times and the ChatGPT router obscuring the true model quality