GPT-6 Sol vs Luna: pricing, workload fit, and what the 50% API cut changes
OpenAI released GPT-6 Sol and Luna with lower API prices. Here is how their costs differ and how to choose between them for agentic versus high-volume workflows.
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026 with a clear split: Sol is aimed at complex coding and agentic workflows, while Luna is the efficiency model for focused, high-volume work. The price gap is large enough that choosing the model by workload can matter more than simply defaulting to the more capable option.
GPT-6 Sol vs Luna pricing
Standard API pricing for prompts up to 272K input tokens
| Model | Input / 1M | Cached input / 1M | Output / 1M | OpenAI positioning |
|---|---|---|---|---|
| GPT-6 Sol | $2.00 | $0.20 | $10.00 | Complex coding and agentic workflows |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | Focused, high-volume tasks |
At Standard rates, Luna costs one twentieth of Sol for both uncached input and output tokens. That does not mean Luna is the better model for every task; it means the cost of routing simple repeatable work to Sol can compound quickly at scale.
What that looks like in a practical workload
For a workload using 100 million uncached input tokens and 20 million output tokens, Sol would cost about $400 at the listed Standard short-context rates: $200 input + $200 output. Luna would cost about $20: $10 input + $10 output. This is a pricing calculation, not a claim that both models would produce equivalent results on the workload.
When Sol makes more sense
- The task involves complex coding, tool use or multi-step agentic execution.
- A failed or weak result is expensive enough that capability matters more than token cost.
- The workflow requires stronger reasoning across a changing sequence of steps.
- You are using AI for lower-volume work where model cost is a small part of the total value created.
When Luna is the stronger economic fit
- The task is focused, repeatable and high-volume.
- You can validate outputs automatically or cheaply.
- The workflow processes large numbers of similar records, classifications, transformations or first-pass generations.
- You can escalate difficult cases to Sol instead of running every request on Sol.
The useful pattern is not Sol versus Luna as a permanent choice. It is model routing. Start by separating tasks by complexity and cost of failure. Use the cheaper model where the workflow is constrained and verifiable, then escalate the smaller set of difficult cases to the stronger model. That can preserve capability where it matters while keeping high-volume automation economical.
A simple routing rule
Route by workload, not model hype
| Workload | Default starting point | Why |
|---|---|---|
| Complex coding or agentic execution | GPT-6 Sol | Matches OpenAI’s stated positioning for complex coding and agents. |
| Focused bulk processing | GPT-6 Luna | Matches OpenAI’s stated high-volume efficiency positioning. |
| Mixed workload | Luna then escalate | Use Luna for constrained cases and route exceptions to Sol after evaluation. |
| High-cost failure cases | Evaluate Sol first | Token savings may be less important than reliability for the specific task. |
Pricing caveats that matter
- The headline rates above are Standard prices for prompts with up to 272K input tokens.
- Longer prompts use higher rates; the current pricing page lists different long-context pricing.
- Cached input is cheaper than uncached input, so real workload cost depends on cache behavior.
- Batch, Flex, Fast mode and regional processing can change the effective price.
- Cost per token is not the same as cost per successful task. Benchmark your own workflow before moving production traffic.
GPT-6 Luna changes the economics of repeatable AI work because its Standard token rates are 20x lower than Sol’s short-context rates. Sol remains the model OpenAI positions for complex coding and agentic workflows. For many production systems, the practical answer is to route work between them rather than choose only one.
Sources & useful resources
- OpenAI API changelog— Official September 22 release and standard pricing.
- OpenAI model catalog— Official positioning and model specifications.
- OpenAI API pricing— Current standard pricing and long-context tier details.