GPT-6 Luna
The cheap end of GPT-6. Fast, and not a substitute for Sol or Opus.
32Index · High
$0.03Per measured task
1067 pts/$Points / $
123Tokens / sec
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 37 | $0.07 | 127 |
| Extra high | 34 | $0.04 | 131 |
| High | 32 | $0.03 | 123 |
| Medium | 29 | $0.02 | — |
| Low | 21 | $0.0045 | 134 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
GPT-6 Luna
Cache appliedSpecs
- API id
- gpt-6-luna
- Released
- 22 Sep 2026
- Context
- 1M
- Max output
- —
- Input / 1M
- $0.10
- Output / 1M
- $0.50
- Cache read / 1M
- $0.01
- Knowledge
- —
- Inputs
- Text
- Weights
- Closed
OpenAI's volume model. Above 272k input tokens the whole request is repriced. Context figure is the 1M listed by Artificial Analysis.
Vision is not confirmed on the sources used for this card.
Use it when
Classification, extraction, routing, and other calls you make millions of. Max effort reaches 37 on the index for about $0.07 a measured task.
Leave it when
Frontier coding or long agent work. Luna is a different product from Sol.