Codex model pricing in AI credits
Every current-generation Codex model, ranked by the credits a typical daily-coding request burns. Verified 2026-08-03.
| Model | Category | Credits / 1M in | Credits / 1M out | Credits / request |
|---|---|---|---|---|
| GPT-5.6 Luna | Lightweight | 5.00 | 30.00 | 0.1 |
| GPT-5.4 mini | Lightweight | 18.75 | 113 | 0.3 |
| GPT-5.6 Terra | Versatile | 50.00 | 300 | 0.8 |
| GPT-5.6 Sol | Powerful | 125 | 750 | 1.9 |
Daily-coding request: a defined workload basket, not a measured task, priced against OpenAI's published credit rate card. See each model's page for the worked cost and the direct-API comparison.
Image generation
GPT-Image-2 sits on the same credit rate card, priced per 1M tokens. Its tokens are image tokens, so these rates are not comparable to the coding models above and the calculator leaves them out.
| Model | Credits / 1M in | Credits / 1M cached | Credits / 1M out | API $ / 1M in |
|---|---|---|---|---|
| GPT-Image-2 (image) | 200 | 50.00 | 750 | $8.00 |
| GPT-Image-2 (text) | 125 | 31.25 | 250 | $5.00 |
OpenAI's API card publishes no text OUTPUT rate for GPT-Image-2, but the credit card prices text output at 250 credits per 1M. That is recorded as published rather than filled in from the other card.