I primarily use Claude Code (cc) for development, but the Pro membership’s quota is surprisingly limited. I frequently hit the ceiling and have to wait for my weekly or hourly reset. To keep my momentum, I needed a reliable backup. While DeepSeek is cheap, it doesn’t quite match the quality I need. I decided to try Codex, and it has exceeded my expectations. Since they are products from different companies, there are some nuances to using them together. Here’s a summary of my experience.
(Note: This post is continuously updated.)
Requirements
- Codex: Requires a ChatGPT Plus (or higher) subscription; it is not available for free users.
- Claude Code: Requires a Pro or Max membership, or a pay-as-you-go API account.
Project Configuration Files
Codex uses the /init command to generate an AGENTS.md file, while Claude Code uses /init to generate a CLAUDE.md file. Maintaining both manually is tedious.
My solution is to link them: in AGENTS.md, I added the instruction: Read CLAUDE.md for more details on development commands, project architecture, and key directories. My testing shows that Codex will then automatically read the contents of CLAUDE.md. This allows me to maintain only the CLAUDE.md file for project-specific rules. You can also do the reverse if you prefer AGENTS.md.
Custom Commands
Custom slash commands are implemented differently in Claude Code and Codex. Unfortunately, these cannot be shared directly and must be configured separately for each CLI. If you don’t use custom commands, you can ignore this part.
Final Thoughts
- CLI Experience: In terms of pure terminal user experience, Claude Code (cc) is the clear winner. For example,
cchandles session recovery gracefully across logins. Codex requires you to manually select a session upon startup, and it doesn’t effectively isolate sessions by repository, which can lead to a cluttered session list. - Code Quality: Both have their strengths. Codex noticeably outperforms Claude Code when generating code from Figma via the MCP server. However, for general logic and iterative coding, I still find Claude’s output to be slightly superior.
- Cost vs. Quota: Given the quota issues, I’ve opted for a dual-subscription strategy: GPT Plus + Claude Pro. Since both memberships also provide access to their respective web interfaces, I find the value proposition to be solid. If you’re constantly hitting your Claude quota, this hybrid approach is a great way to stay productive.
