GPT-5-Codex: Agentic Coding with Layered Safety

GPT-5-Codex is a GPT-5 variant tuned for agentic coding, trained via reinforcement learning on real-world tasks to produce human-like, instruction-precise code and to self-test until passing. It’s available through local CLI/IDE tools and cloud platforms including Codex web, GitHub, and ChatGPT mobile. The addendum emphasizes comprehensive safety measures, from specialized model training to sandboxing and configurable network access.

Key Points

• GPT-5-Codex is a GPT-5 variant optimized for agentic coding tasks in Codex.
• It uses reinforcement learning on real-world coding tasks to produce human-like, instruction-faithful code and to iteratively run tests until passing.
• Availability spans local (CLI and IDE extensions) and cloud (Codex web, GitHub, ChatGPT mobile) environments.
• The addendum outlines comprehensive safety measures at both the model and product levels.
• Mitigations include specialized safety training, prompt-injection defenses, agent sandboxing, and configurable network access.

Sentiment

The community is generally enthusiastic about GPT-5's coding capabilities, with many users reporting switching from Claude Code. However, the discussion reveals significant frustration with Codex's product-level issues including rate limits, UX quirks, stability, and lack of edit approval workflows. Almost no one discusses the safety addendum itself, treating the thread purely as a product comparison. The overall tone leans positive toward GPT-5 as a model while acknowledging the tooling still needs work.

In Agreement

• GPT-5 writes clean, minimalist code that follows instructions well and handles long context without leaving tasks half-done or mocking functions
• GPT-5 shows substantial improvement in tool calling — knowing when and what to search for when navigating codebases
• Codex CLI is getting constant, meaningful updates and is more steerable than Claude Code, with more generous usage limits
• GPT-5's refactoring capabilities are notably improved, with the model being praised for producing artful, human-like code

Opposed

• Codex stops frequently to ask permission rather than working to completion, even with explicit instructions to continue, and suffers severe context degradation near max window
• The Codex CLI Rust rewrite has poor UX compared to Claude Code's terminal interface, and rate limiting is aggressive for API users making it near-unusable without a subscription
• No granular edit approval mode exists — Codex either autonomously edits everything or gives diffs to copy-paste, lacking Claude Code's interactive approval workflow
• GPT-5-Codex thinking time is excessively long, stability issues persist with frequent downtime, and privacy concerns remain about sending proprietary code to OpenAI without a privacy mode