Codex Pricing & Usage Limits Explained: Plans, Resets, and Capacity Errors

Codex is the coding assistant integrated across all ChatGPT subscription tiers — from Free to Enterprise. But what tends to trip developers up isn't the features themselves; it's the usage quotas, reset cycles, and that maddening "Selected model is at capacity" message. This article breaks down Codex's current pricing structure and quota logic so you can pick the right plan and make the most of your allowance.

Codex Plans and Pricing Structure

Codex is available across the full ChatGPT subscription lineup, with each tier offering a different usage capacity and request priority.

Here's the breakdown:

• Free: Codex is accessible on the free tier, but with limited quota — suitable for light experimentation.
• Plus: $20/month, the go-to entry point for most individual developers.
• Pro ($100 tier): $100/month, offering roughly 5× the usage of Plus.
• Pro ($200 tier, 20×): $200/month, offering approximately 20× the capacity of Plus. Note that new sign-ups for this tier have been paused as of September 10, 2026.
• Enterprise: Designed for teams, with higher capacity and administrative controls.

The pricing ladder reflects a clear philosophy from OpenAI: pay more, get more quota and higher request priority — not entirely new features. For developers who rely heavily on Codex for code generation, refactoring, and debugging, the Pro tier offers noticeably better value.

Codex is OpenAI's LLM-powered coding assistant, purpose-built for software development tasks. It's integrated as a feature module within the ChatGPT interface rather than a standalone API product (OpenAI shut down the independent Codex API in 2023). It handles code generation, completion, refactoring, debugging, and natural language-to-code tasks, all powered by the GPT-4 model family. Because code generation demands significantly more compute than typical chat interactions — longer context windows, heavier inference — OpenAI manages Codex capacity independently from standard ChatGPT conversations. This is why Codex has its own separate quota system rather than simply sharing ChatGPT's message limits.

How Limits Are Calculated and Reset

Codex quotas aren't a simple monthly total. Instead, they use a two-layer limit system.

The 5-Hour Rolling Window

The core limit is measured within a rolling 5-hour window. Usage within any given 5-hour interval is tracked separately, and the quota resets when the window closes. This design prevents users from burning through all their resources in a short burst, helping ensure overall service stability and availability.

A rolling window is a common rate-limiting design pattern, and it works differently from a fixed window. A fixed window resets at a set clock time — say, on the hour. A rolling window starts from the moment of your first request and rolls forward from there. This means you can't game the system by waiting for a clock reset — each call re-anchors the window start. Rolling windows are friendlier to the server side, effectively preventing the "thundering herd problem" where large numbers of users flood the system at the same moment. The trade-off for users is that it's harder to predict exactly when your quota will refresh — you need to remember when you last made a heavy burst of requests.

Weekly Cap on Top

Layered on top of the 5-hour window is a weekly total cap. Even if you pace yourself within each 5-hour window, sustained high-volume usage can still hit the weekly ceiling. This combination of a short rolling window plus a weekly cap is designed to smooth out usage curves and prevent extreme load spikes.

For developers, this is an important detail: if you're heading into a project sprint that requires heavy Codex usage, it's worth planning your workload in advance — spread intensive sessions across different time windows rather than front-loading everything into one or two days.

What to Do When You See "Selected model is at capacity"

"Selected model is at capacity" is one of the most common errors Codex users encounter. This message typically doesn't mean your account quota is exhausted — it means the model's server-side load is maxed out at that moment.

Here are a few ways to handle it:

• Retry after a few minutes: Capacity issues are often temporary. Once server load eases, requests usually go through normally.
• Switch models: If the model you're using stays congested, switching to another available model may restore access immediately.
• Avoid peak hours: Usage pressure drops significantly if you work around high-traffic periods.
• Check your quota status: Distinguish between "at capacity" (a server-side issue) and "quota exhausted" (an account-level issue) — the fixes are entirely different.

It's worth keeping these two issues separate. A capacity error is a bottleneck in OpenAI's infrastructure at a specific moment. A quota error is the boundary of your personal allowance. Understanding the difference helps you diagnose and resolve the problem faster.

Choosing the Right Plan

Given the pricing and quota logic, choosing a plan is fairly straightforward:

• Light users: Free or Plus is enough. The $20 Plus plan handles everyday small-scale coding assistance without issue.
• Professional developers: If Codex has become a core part of your daily workflow, the $100 Pro plan's 5× capacity will meaningfully reduce "quota anxiety."
• Heavy and team users: The $200 20× tier offers ample capacity, but new sign-ups are paused — keep an eye on when it reopens. For team use cases, Enterprise is the direct path.

Overall, Codex's pricing and quota system reflects OpenAI's ongoing balancing act between open access and resource control. Rather than bumping into limits or capacity walls reactively, it pays to understand the rules upfront and direct your quota where it matters most.