Rate Limits · Overview
Codex Rate Limits by ChatGPT Plan
Both Plus and Pro include Codex access. A limited-time 2× rate boost is currently active, so the figures below are the boosted rates (they vary with task complexity and context). Data updated July 2026.
ChatGPT Plus
Basic Codex access
- GPT-5.6 series / 5h: 15-280
- GPT-5.5 / 5h: 15-80
- GPT-5.4 series / 5h: 20-350
- Cloud Tasks / 5h: Unavailable
- Code review / 5h: Unavailable
ChatGPT Pro
Full-power Codex access
- GPT-5.6 series / 5h: 75-5600 (5x-20x)
- GPT-5.5 / 5h: 75-1600 (5x-20x)
- GPT-5.4 series / 5h: 100-7000 (5x-20x)
- Cloud Tasks / 5h: Unavailable
- Code review / 5h: Unavailable
Detailed model rate comparison
All limits refresh on a 5-hour cycle.
| Model | Plus | Pro (5x) | Pro (20x) | Business | API |
|---|---|---|---|---|---|
| GPT-5.6 Sol (Local messages / 5h) | 15-90 | 75-450 | 300-1800 | 15-90 | Pay as you go |
| GPT-5.6 Terra (Local messages / 5h) | 20-110 | 100-550 | 400-2200 | 20-110 | Pay as you go |
| GPT-5.6 Luna (Local messages / 5h) | 50-280 | 250-1400 | 1000-5600 | 50-280 | Pay as you go |
| GPT-5.5 (Local messages / 5h) | 15-80 | 75-400 | 300-1600 | 15-80 | Pay as you go |
| GPT-5.4 (Local messages / 5h) | 20-100 | 100-500 | 400-2000 | 20-100 | Pay as you go |
| GPT-5.4 mini (Local messages / 5h) | 60-350 | 300-1750 | 1200-7000 | 60-350 | Pay as you go |
Key mechanics explained
Understand how limits are calculated and what to do when you run out.
Shared local + cloud window
Local Messages and Cloud Tasks share the same 5-hour window, so heavy Cloud Task usage also consumes your local message quota. Additional network-related limits may also apply.
Still on the legacy rate card
As of July 2026, Plus / Pro users are still on the legacy rate card. Some users will gradually be migrated to the new token-based billing model.
What happens when I run out?
Once Plus or Pro limits are exhausted, you can buy additional credits directly — no need to upgrade your plan. You can also switch to the GPT-5.4-mini model to make your quota last longer.
Rate data is for reference only. Actual consumption depends on task complexity — always refer to OpenAI's official documentation.