GPT-5.6 Luna in OpenAI Codex
OpenAI - Lightweight
In OpenAI Codex, GPT-5.6 Luna meters at 25 credits per 1M input tokens and 150 per 1M output tokens. At $0.04 per credit that is exactly $1.00 and $6.00 per 1M tokens, the same as calling the OpenAI API directly. A typical daily-coding request works out to about 0.4 credits (about $0.02). Verified against OpenAI's Codex rate card on 2026-07-21.
Credit rate card
- Credits / 1M in
- 25
- Credits / 1M cached
- 2.5
- Credits / 1M out
- 150
- Cost per credit
- $0.04
Credits per request
Defined workload baskets, not measured tasks.
| Workload | Credits | Cost | Direct API cost |
|---|---|---|---|
| Short chat turn | 0.2 | $0.01 | $0.01 |
| Daily coding request | 0.4 | $0.02 | $0.02 |
| Agent session step | 1.3 | $0.05 | $0.05 |
At about 0.4 credits per daily-coding request, GPT-5.6 Luna ranks #2 of 4 current-generation models by cost, about 1.3x the cheapest (GPT-5.4 mini).
Credit-to-API parity
The implied dollar value of a credit spent on GPT-5.6 Luna, backed out from the direct API rate card divided by the credit rate card. It should land on $0.04 for both input and output.
- Implied $ / credit, input
- $0.04
- Implied $ / credit, output
- $0.04
- Markup at the daily workload
- 0.00%
- Ratio match
- Exact
Input : cached : output ratio in credits is 10 : 1 : 60, versus 10 : 1 : 60 in direct API dollars.