OpenAI Pricing for GPT-6 Sol and Luna per Million
Sol, the larger of the two GPT-6 models, is $2 per million input tokens and $10 per million output tokens. Luna is $0.10 and $0.50. Both figures are standard-mode rates on OpenAI's developer pricing page, read on October 22, 2026, for requests of up to 272,000 input tokens. OpenAI released the models on September 22, 2026. According to VentureBeat's launch report, an OpenAI spokesperson called the rates permanent rather than promotional, and the company states a 90% discount on cached input.
Standard mode
USD per million tokens. API names: gpt-6-sol, gpt-6-luna.
| Model, prompt size | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-sol, up to 272K | $2.00 | $0.20 | $10.00 |
| gpt-6-sol, above 272K | $4.00 | $0.40 | $15.00 |
| gpt-6-luna, up to 272K | $0.10 | $0.01 | $0.50 |
| gpt-6-luna, above 272K | $0.20 | $0.02 | $0.75 |
Source: OpenAI API pricing, as of October 22, 2026.
Past 272K input tokens the whole price list steps up: Sol input doubles, output rises by half. Check how long your longest prompts really are before assuming the lower row applies to all traffic.
Batch, Flex and Fast
The same pricing page has two more tables. Batch and Flex come to half of standard for these models; Fast is double.
| Mode (up to 272K) | Sol in / out | Luna in / out |
|---|---|---|
| Batch and Flex | $1.00 / $5.00 | $0.05 / $0.25 |
| Standard | $2.00 / $10.00 | $0.10 / $0.50 |
| Fast | $4.00 / $20.00 | $0.20 / $1.00 |
Source: OpenAI API pricing, as of October 22, 2026. The page gives no context window for either model.
A request, line by line
A prompt of 20,000 input tokens, 16,000 of them a cached prefix, and an answer of 1,000 tokens. Standard mode, rates from the first table (OpenAI pricing page, October 22, 2026).
| Line | Sol | Luna |
|---|---|---|
| 4,000 fresh input tokens | $0.0080 | $0.0004 |
| 16,000 cached input tokens | $0.0032 | $0.00016 |
| 1,000 output tokens | $0.0100 | $0.0005 |
| Total | $0.0212 | $0.00106 |
On Sol the output line is almost half the bill, from one twentieth of the tokens. Without caching the input alone would be $0.040 and the total $0.050, so the cached prefix saves more than half.
Next to the 2023 rates
The calculator on the site's home page shows April 2023 rates: GPT-4 with the 8K context at $0.03 per 1K tokens, GPT-3.5 Turbo at $0.002. Multiply by a thousand and those are $30 and $2 per million. Sol's $2 input equals the old GPT-3.5 Turbo rate and is a fifteenth of the old GPT-4 rate; Luna's $0.10 is a twentieth of the GPT-3.5 Turbo rate.
Treat that as a price comparison only. The 2023 tool shows a single rate per model with no separate output charge, and the models are different products. When a figure here disagrees with OpenAI's page, the page is right. The docs page shows how the 2023 calculator turns a prompt into a cost.