GPT-6 API Pricing: Astra vs Sol vs Luna, and What It Costs

Jonhisking · October 2, 2026

OpenAI released three GPT-6 models in September 2026: GPT-6 Astra, the flagship, on September 4, and GPT-6 Sol and GPT-6 Luna on September 22. Their API prices are very different, so picking the right one matters far more than any discount. Here are the prices, what they mean for real workloads, and how they compare with Claude and Gemini.

GPT-6 API prices

Per million tokens:

Astra is 5 times the price of Sol, and Sol is 20 times the price of Luna. That is a 100× range from the cheapest to the most expensive model in the same family.

Requests with more than 272,000 input tokens are billed at a higher rate, so very long contexts cost more than this table suggests.

How that compares with GPT-5.6

Sol and Luna launched at half the price of the models they replace. GPT-5.6 Sol was $4 input and $20 output; GPT-5.6 Luna was $0.20 and $1.20. If you are still on GPT-5.6, switching model names alone halves the bill.

What it costs in practice

A typical request with 2,000 input tokens and 500 output tokens:

Prices for the other models are list API prices checked October 1, 2026.

GPT-6 Sol costs exactly the same per token as Claude Sonnet 5.5. GPT-6 Astra costs 2.5 times as much as Claude Opus 5.5.

Which GPT-6 model should you use?

GPT-6 Luna for high-volume, simple work: classification, extraction, routing, short answers, summaries of short texts. At $0.10 per million input tokens it is cheap enough to run on every request.

GPT-6 Sol as the default for most apps and coding work: chat assistants, writing, analysis, agents. It is the sensible middle.

GPT-6 Astra only where you can measure that it does better: long, multi-step agent work, hard reasoning, tasks where a mistake is expensive. Route only those requests to it.

A common setup is to send everything to Sol or Luna by default and escalate to Astra when a cheaper model fails or the task is flagged as hard.

Ways to pay less

1. Use prompt caching. Cached input is a tenth of the normal price on all three models. Put fixed instructions and documents at the start of the prompt. See Prompt caching explained.
2. Use the Batch API for work that can wait, typically at half price. See Batch APIs.
3. Keep output short. Output costs 5 times input on every GPT-6 model.
4. Watch reasoning. Hidden reasoning tokens are billed as output. Use the lowest reasoning effort that works. See Reasoning models.
5. Write prompts in English. On the GPT tokenizer, the same prompt uses about 1.44× the tokens in Korean and 1.79× in Japanese.

GPT-6 in ChatGPT plans

You do not need the API to use GPT-6. GPT-6 Astra is available to ChatGPT Pro, Business Premium and Enterprise users, and is rolling out to Plus, first in ChatGPT's Work section and Codex. For heavy use, a plan can be much cheaper than paying Astra's API rate. Compare in the Subscription vs API calculator.

Calculate your own cost

Paste a typical prompt into the token counter to see its exact token count on GPT models and its cost on every GPT-6 model, Claude and Gemini side by side.

Prices change. Check OpenAI's pricing page before a large job.