OpenAI radically changed how Codex is billed on April 2, 2026. Moving away from the old per-message pricing, the GPT-5.3-Codex model now uses a credit-based system mapped directly to API token usage for Plus, Pro, Business, and Enterprise plans. Input tokens cost 43.75 credits per 1M, output tokens cost 350 credits per 1M, and cached inputs get a massive 90% discount. Here is the complete breakdown of the new GPT-5.3-Codex rate card and how it impacts your subscription.
The April 2026 OpenAI Codex Pricing Overhaul
If you are using the OpenAI Codex CLI or running coding agents through your ChatGPT subscription, the rules of the game have changed. Prior to April 2026, ChatGPT Plus and Pro users simply had a "message limit" when using Codex. A message containing a 50-line Python script cost you exactly the same as a message containing a 15,000-line React monorepo.
That is no longer the case. OpenAI has transitioned Codex usage to a token-based credit system. This aligns consumer and business subscription billing with the reality of how developer APIs work. The bigger your codebase context, the faster you burn through your monthly credits.
The GPT-5.3-Codex Rate Card
The GPT-5.3-Codex model is a highly capable, coding-optimized variant that serves as a faster, cheaper alternative to the flagship GPT-5.4. Here is exactly how OpenAI charges for it under the new credit system:
| Token Type | Cost (per 1 Million Tokens) | Note |
|---|---|---|
| Standard Input Tokens | 43.75 credits | Reading your code and prompts |
| Cached Input Tokens | 4.375 credits | Massive 90% discount for repeated context |
| Output Tokens | 350.00 credits | Writing new code (8x more expensive than input) |
For developers directly hitting the OpenAI API, third-party trackers estimate the historical dollar-equivalent value of this model to be roughly $1.75 per 1M input tokens and $14.00 per 1M output tokens. However, if you are using Codex via a ChatGPT plan, you must track credits, not dollars.
Understanding the 90% Prompt Caching Discount
The biggest saving grace in the new GPT-5.3-Codex pricing model is Prompt Caching. At just 4.375 credits per 1M tokens, cached inputs are 90% cheaper than standard inputs.
If you are using the Codex CLI to build a feature, the first time you ask a question, Codex reads your entire workspace (e.g., 200,000 tokens). This costs 8.75 credits. But when you ask your second question 30 seconds later, Codex doesn't need to re-read the workspace from scratch. It uses the cached context, dropping the cost of reading that 200,000 tokens to less than 1 credit.
Pro Tip: Do not clear your Codex CLI terminal session frequently if you are working on the same files. Keeping the session open keeps the cache warm, saving you 90% on input token credits.
How Does This Compare to GPT-5.4?
GPT-5.3-Codex is explicitly positioned as the workhorse for high-volume coding (we tested this extensively in our 30-day coding assistant review). It is significantly cheaper than the flagship GPT-5.4 model. If you use GPT-5.4 for agentic coding, not only is the base rate higher, but you also risk hitting the infamous 2x billing surcharge for context over 272K tokens.
For 90% of daily programming tasks—like generating boilerplate, writing unit tests, or debugging a specific function—GPT-5.3-Codex offers the best balance of speed and credit efficiency.
Subscription Limits: Plus vs. Pro vs. Business
Since Codex usage now burns credits, your subscription tier dictates how much coding you can actually do before hitting a wall:
- ChatGPT Plus ($20/mo): You get a baseline monthly credit allowance. If you use Codex to ingest large monorepos daily, you will likely run out of credits before the month ends.
- ChatGPT Pro ($100/mo): Designed specifically for power users, the $100 Pro tier offers five times the Codex usage limits of the Plus plan. If you are a full-time software engineer relying on Codex, this is the minimum tier you need.
- ChatGPT Business/Enterprise: As of April 2026, Business workspaces can mix standard seats (fixed monthly cost) and dedicated Codex seats (usage-based, no fixed cost). This allows teams to only pay for the exact amount of GPT-5.3-Codex credits their developers consume.
3 Ways to Optimize Your Codex Credit Usage
If you want your monthly ChatGPT subscription credits to last longer, you need to manage your token consumption:
- Limit Output Tokens: Output tokens (350 credits/1M) are 8x more expensive than input tokens. Stop asking Codex to "rewrite the entire file." Instead, ask it to "show me the 5 lines I need to change."
- Configure .codexignore: Prevent the AI from reading large, useless files like
package-lock.jsonor compileddist/folders. Every file it reads burns input tokens. - Leverage the Cache: Keep your sessions alive while working on the same problem. A warm cache applies a 90% discount to all input tokens in that session.
Frequently Asked Questions
What is the pricing for GPT-5.3-Codex?
As of April 2026, OpenAI charges 43.75 credits per 1M input tokens, 4.375 credits per 1M cached input tokens, and 350 credits per 1M output tokens for GPT-5.3-Codex via ChatGPT plans.
Why did OpenAI change Codex pricing from per-message to tokens?
OpenAI transitioned to a token-based credit system to account for the massive variance in codebase sizes. A developer uploading a 200,000-token codebase consumes vastly more compute than someone uploading a 500-token script, and the new credit system reflects this.
Does prompt caching apply to Codex?
Yes. Cached input tokens in GPT-5.3-Codex are billed at just 4.375 credits per 1M tokens, which is a 90% discount compared to standard input tokens. Keeping your Codex CLI session open keeps the cache warm.
Is GPT-5.3-Codex included in ChatGPT Plus?
Yes, GPT-5.3-Codex is available to ChatGPT Plus subscribers, but usage consumes your monthly credit allowance. Heavy users may need to upgrade to the $100/month Pro plan, which offers 5x the Codex usage limit.