GPT-5.4 mini in Codex: 2x Faster, 30% Cheaper
GPT-5.4 mini is now live in Codex. It's 2x faster, uses 70% fewer tokens, and improves over GPT-5 mini on coding and reasoning. Perfect for codebase exploration and subagent work.
TL;DR
- GPT-5.4 mini now available in Codex — 2x faster than GPT-5.4, uses 30% of the token budget
- Better at coding, reasoning, image understanding, and tool use than GPT-5 mini
- Use it for codebase exploration, file review, and subagent work; stick with GPT-5.4 for complex planning
New
- GPT-5.4 mini model — Fast, efficient model for lighter coding tasks that runs 2x faster and costs 70% less in token usage, letting comparable tasks run 3.3x longer before hitting limits.
- Multi-platform availability — Available in Codex app, CLI, IDE extension, web, and API.
How to Switch
- CLI: Start a new thread with
codex --model gpt-5.4-minior use/modelduring a session. - IDE extension: Select GPT-5.4 mini from the model selector in the composer.
- Codex app: Choose GPT-5.4 mini from the model selector in the composer.
When to Use It
- Codebase exploration and large-file review
- Processing supporting documents
- Subagent work and less reasoning-intensive tasks
- Reserve GPT-5.4 for complex planning, coordination, and final judgment calls
Update your CLI, IDE extension, or Codex app to the latest version if GPT-5.4 mini doesn't appear yet.
Source: Codex
