---
title: "GPT-6 API Pricing: Sol and Luna Rates for Cache, Batch, and Fast"
description: "OpenAI released gpt-6-sol and gpt-6-luna on September 22. Official prices, cache and long-context billing, and a sample monthly cost for each tier."
url: https://cdnping.org/en/blog/gpt-6-sol-luna-api-price
language: en
published: 2026-10-19
category: "Model watch"
tags: ["OpenAI","GPT-6","API pricing","cached input","Batch"]
source: cdnping.org
license: Free to quote and summarize with a link to the url above.
---
# GPT-6 API Pricing: Sol and Luna Rates for Cache, Batch, and Fast

OpenAI released gpt-6-sol and gpt-6-luna in its API changelog on September 22. According to the official pricing page, GPT-6 API pricing for inputs up to 272K tokens works out like this: Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 input and $0.50 output. Cached input is one tenth of the input price for both. Batch and Flex are half the standard price, and Fast is double.

All prices below come from OpenAI's official pricing page and are as of October 2026. Treat the [official pricing page](https://developers.openai.com/api/docs/pricing) as the source of truth.

## What are these two models?

Per the [official changelog](https://developers.openai.com/api/docs/changelog), both are reasoning models. They accept text and image input, produce text output, and can be called through either the Responses API or the Chat Completions API. Sol is the more expensive tier; Luna's unit price is roughly one twentieth of Sol's. For the capability differences, see OpenAI's model documentation. This article covers pricing only.

## Price table: what does a million tokens cost?

For inputs up to 272K tokens (USD per million tokens):

| Model and tier | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-sol standard | 2.00 | 0.20 | 10.00 |
| gpt-6-sol Batch / Flex | 1.00 | 0.10 | 5.00 |
| gpt-6-sol Fast | 4.00 | 0.40 | 20.00 |
| gpt-6-luna standard | 0.10 | 0.01 | 0.50 |
| gpt-6-luna Batch / Flex | 0.05 | 0.005 | 0.25 |
| gpt-6-luna Fast | 0.20 | 0.02 | 1.00 |

A few notes:

- Fast is the new name for what used to be Priority; the rename happened on July 30. The changelog says requests that carry the `priority` flag are automatically handled as Fast, so you don't need to change your code.
- Regional processing (data residency) endpoints add a 10% surcharge. The pricing page states this for models released on or after March 5, 2026. It doesn't list release dates for these two models separately, so confirm with OpenAI before relying on it.

## How are inputs over 272K billed?

Requests with more than 272K input tokens are billed at long-context prices:

| Model and tier | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-sol standard | 4.00 | 0.40 | 15.00 |
| gpt-6-luna standard | 0.20 | 0.02 | 0.75 |

Input doubles and output is 1.5x. How Batch, Flex, and Fast are priced at long context is covered only by the official pricing page.

The pricing page says requests with more than 272K input tokens are billed at the long-context price. Whether the whole request is billed that way is something to confirm in the official documentation. Before you split a long document into several requests, work out whether it actually saves money.

## How much do caching and Batch save? (Example)

The following is a hypothetical workload, not a measurement: in one month, 100 million input tokens and 20 million output tokens, with 60% of input hitting the cache and every request under 272K.

| Tier | Sol | Luna |
|---|---|---|
| Standard | $292 | $14.6 |
| Batch / Flex | $146 | $7.3 |
| Fast | $584 | $29.2 |

Take Sol standard as an example: 40 million uncached input tokens cost $80, 60 million cached tokens cost $12, and output costs $200, for a total of $292. In this scenario output dominates the bill, and caching only reduces the input portion. So caching pays off most for tasks with long prompts and short outputs.

## When should you pick Batch, Flex, or Fast?

The pricing page lists prices only. Limits for each tier (latency, availability, rate limits) are in the official documentation. As a rule of thumb:

1. Offline evals, bulk labeling, overnight jobs: nothing needs to come back in real time, so look at Batch or Flex first. They cost half as much.
2. Interactive products where a user is waiting: use the standard tier. Consider Fast only when response speed directly affects the experience, since it costs double. The changelog notes that Fast mode for the earlier GPT-5.6 Sol could be up to 2.5x faster than standard; for the actual GPT-6 speedup, see OpenAI's documentation.
3. Simple classification, rewriting, extraction: try Luna first, and move to Sol only if the quality isn't good enough.
4. Reusing the same long prefix: put the fixed content at the start of the prompt to raise cache hits, and check your bill to confirm the cache is actually working.

## Is the "API prices cut 50%" claim reliable?

A third-party tracking page records that someone posted on September 22 mentioning "API prices cut 50%". I couldn't find that wording in the official changelog or pricing page, and I couldn't check the original post, so I'm not drawing a conclusion. The only thing confirmed on the official pages is the price table above. Batch and Flex being 50% below standard is a relationship between tiers, not a price cut for the model overall.

## Do you need to act on the September 25 image encoding fix?

Yes. The September 25 changelog entry says a bug in image encoding, which had degraded image understanding for GPT-6 Sol and Luna, was fixed. OpenAI recommends rerunning evals that use image input and retrying affected workflows. If you ran any image-related evaluation between September 22 and 25, your conclusions may need to be redone.

## How do you keep watching prices and availability?

- OpenAI's pricing page and changelog change over time. [/en/changes](/en/changes) shows line-by-line change records for OpenAI and other official pages.
- For availability and latency after a model launches, see [/en/models](/en/models).
- The full workflow is in [How to find out right away when LLM API prices change](/en/blog/track-llm-price-and-model-changes).
- If your bill and token counts don't match, see [How to debug wrong API billing](/en/blog/relay-billing-token-mismatch).
- To compare the cost structure of different access routes, see [How to choose an AI API access route](/en/blog/official-api-vs-cloud-vs-gateway).
- For a comparable case, see [how Claude Haiku 5.5 is priced](/en/blog/claude-haiku-5-5-pricing), which also has tiered rules based on context length.

## Sources

- OpenAI API changelog: https://developers.openai.com/api/docs/changelog
- OpenAI API pricing page: https://developers.openai.com/api/docs/pricing