How many credits does it actually cost to generate an image with Codex? I checked the official OpenAI documentation.
What Are "Credits" Anyway?
In the Codex world, credits are the unit of usage for your ChatGPT plan.
Think of it like your mobile data plan. You pay a monthly fee and get a certain amount of data. Different activities use different amounts — streaming video uses more than texting. Same with Codex.
- Plus ($20/month): A fixed number of messages per 5-hour window
- Pro ($100/month): 5x the usage of Plus
- When you run out: Buy more credits, wait for the reset, or switch to a lighter model
Credits reset every 5 hours. Different models and features consume them at different rates. GPT-5.4-mini is light (~1-2 credits per message). Fast mode is heavy (2x consumption). Image generation is what this article is about.
Note: Business and Enterprise plans are moving to a per-token billing model, but Plus/Pro still use the simpler "~X credits per message" system. The numbers in this article use Plus/Pro credit notation.
TL;DR
One 1024×1024 image generation = ~5-6 credits.
A regular text message (GPT-5.4) costs ~7 credits on Plus/Pro plans, so image generation is actually slightly cheaper per operation.
Credit Comparison Table
Image generation is often assumed to be heavier than text, but at the base size it's actually lighter. Higher quality and larger sizes can increase consumption up to 3-5x.
Does the Skill Itself Eat Tokens?
A common concern: "If I keep the imagegen skill installed, does it bloat my context?"
Codex uses progressive disclosure for skills:
So unless you actually ask Codex to generate an image, the skill has almost no impact on your context window.
Is It "Intercepting" the API?
No. Codex's built-in image generation simply uses the Responses API's native image_generation tool. No user API key required — it runs entirely within your ChatGPT plan.
Your prompt
→ GPT-5.5 rewrites it (prompt optimization)
→ Internal gpt-image-2 generates the image
→ Result returned as base64CLI Fallback Is a Different Beast
The CLI fallback charges your own API key at standard image API rates, completely independent of Codex's rate limits.
Codex image generation consumes roughly the same or fewer credits than a regular text message. The skill's context impact is minimal thanks to progressive disclosure. You can ask Codex to "draw something" without worrying about burning through your limits too fast.
