Codex Limits Comparison: Plus / Business / Pro in One Clear Table
1. First, Understand: What Is Codex's "5-Hour Limit"
Codex's quota mechanism is calculated on a per-5-hour window basis. The number of requests (Local Messages) within each window has a cap; once reached, you need to wait for the next window to refresh. Each model's limit is calculated independently — running out of GPT-5.6 Luna doesn't mean GPT-5.4 mini is also exhausted; you can switch models and keep working.
Currently, Codex offers three usage dimensions:
| Dimension | Meaning | Current Status |
|---|---|---|
| Local Messages / 5h | Code generation requests in local terminal or IDE, refreshes every 5 hours | Live — this article's focus |
| Cloud Tasks / 5h | Cloud-based asynchronous tasks | Not yet available |
| Code Reviews / 5h | Automated code review | Not yet available |
All data below revolves around Local Messages / 5h — the one you actually use day to day.
2. One Table: Complete Limits for Four Plans Across Six Models
Data source: OpenAI Codex official usage dashboard (updated April 2026).
| Model | Plus | Business | Pro 5x | Pro 20x |
|---|---|---|---|---|
| GPT-5.6 Sol | 15–90 | 15–90 | 75–450 | 300–1,800 |
| GPT-5.6 Terra | 20–110 | 20–110 | 100–550 | 400–2,200 |
| GPT-5.6 Luna | 50–280 | 50–280 | 250–1,400 | 1,000–5,600 |
| GPT-5.5 | 15–80 | 15–80 | 75–400 | 300–1,600 |
| GPT-5.4 | 20–100 | 20–100 | 100–500 | 400–2,000 |
| GPT-5.4 mini | 60–350 | 60–350 | 300–1,750 | 1,200–7,000 |
Below are actual screenshots of each plan from the OpenAI official usage dashboard:
A few takeaways you can spot immediately:
Plus and Business have exactly the same Codex limits. The numbers across all 6 models show no difference whatsoever. Business's advantage is not in Codex limits but in: 15 bonus Pro model usage sessions (Pro models are otherwise exclusive to the $200/month Pro membership), and team collaboration management features. The price has also dropped from the original $30 to $25.
Pro 5x is roughly 5 times the Plus limit. The actual multiplier ranges from 4.7x to 5.4x depending on the model. Take GPT-5.6 Luna as an example: Plus gets 50–280 messages, Pro 5x gets 250–1,400 messages.
Pro 20x is an absolute order-of-magnitude leap. GPT-5.6 Luna up to 5,600 messages/5h, GPT-5.4 mini up to 7,000 messages/5h. Heavy developers don't need to stare at limit tables and budget carefully — Pro 20x is designed so you can forget about limits entirely.
Small models save usage; large models save time. Within the same plan, switching from GPT-5.6 Sol to GPT-5.4 mini roughly multiplies your usage by 3.3–3.9x. Don't use large models for simple tasks (formatting, autocomplete, comment translation).
3. How to Check Remaining Usage
Four methods, ordered by convenience:
Method 1: Codex App Client (Recommended)
To promote the Codex application, OpenAI offers doubled usage limits when using the App. With the same plan and model, the App allows twice as many requests as the CLI and web versions. Click the local button in the interface to view your current usage.
Method 2: Web Dashboard
Visit chatgpt.com/codex/settings/usage in your browser. After logging in, you'll directly see your current remaining quota and model distribution.
Method 3: CLI Terminal
Type /status in the terminal for instant feedback on your current usage status. Ideal for those who don't want to leave the command line.
Method 4: IDE Editor
In supported IDEs, click the Codex button in the bottom-left corner of the interface — a panel pops up directly showing your remaining usage.
The cost of Fast mode: Enabling Fast mode significantly improves response speed, but doubles your usage consumption. Turn it off for non-urgent tasks.
4. Practical Token-Saving Tips
These four aren't generic advice — they're lessons learned after actually using Codex for a month:
1. Switch to the mini model for simple tasks — extends your usage by roughly 3.3x.
Official data confirms: GPT-5.4 mini's limit is 3.3–3.9x that of GPT-5.6 Sol. Formatting code, generating comments, renaming variables — these don't need a large model. Build the habit: before every task, glance at which model you're currently using.
2. New window = savings.
Codex injects the current session's context into the model. If you have 20 files open and 200 rounds of conversation with Codex — every new request carries all that context. For unrelated historical threads, creating a new session window clears the context, and token consumption drops noticeably.
3. Place AGENTS.md in a subdirectory, not the root.
If you use AGENTS.md in a large project, nest it in a specific subdirectory rather than the project root to control the context size injected each time. Codex by default searches upward from the current file's directory for AGENTS.md.
4. Disable unused MCP servers.
Each additional MCP server adds another layer of context wrapping to every message — it adds up. Disable them promptly when not needed.
But when you hit a core business logic problem — use the strongest model without hesitation. A single deep Pro model task takes about 70–90 minutes to produce results, and the output quality is not something mini can match. The cost of spending two hours stuck on a bug far exceeds the tokens you'd save.
5. How to Buy
The four plans' pricing and purchase channels:
| Plan | Monthly Fee | Best For |
|---|---|---|
| Plus | $20 | Individual developers; daily Codex usage is sufficient |
| Business | $25 (just dropped in price) | Users who want Pro model access + team management |
| Pro 5x | ~$100 | High-intensity Codex users |
| Pro 20x | $200 | Professional developers; use without thinking about limits |
Through upgpts.com, you can upgrade to Plus or Business below the official standard price, no overseas credit card required. Stable operation for over a year.
More related tutorials:
- GPT-5.6 Usage Guide: How to Choose Sol, Terra, Luna? Upgrade Tutorial for Users in China
- ChatGPT Plus vs Pro Plan Comparison (Latest 2026)
- ChatGPT Business vs Plus: Which One to Choose?