GPT-6 API pricing: Astra, Sol and Luna cost per million tokens

GPT-6 Astra, GPT-6 Sol and GPT-6 Luna are billed per token on SeedRouter, with separate rates for input, output, cached input and cache writes. GPT-6 Luna is the cheapest, GPT-6 Astra the most expensive, and GPT-6 Sol sits between them. A prompt over 272K input tokens is billed at a higher long-context rate for the whole request. There is no subscription, and a request that fails is not charged.

The tables below are read from the rate table that bills your account, so they show what a request costs today.

What do GPT-6 Astra, Sol and Luna cost per million tokens?

Each table has two rows: prompts up to 272K input tokens, and prompts over it. Output includes the reasoning tokens the model spends before it answers.

What are OpenAI's list prices?

For reference, this is how OpenAI's Standard prices are structured, from its pricing page. Each line is a multiple of that model's own input rate, and the three models share the same shape:

Across models, GPT-6 Astra's input rate is five times GPT-6 Sol's, and GPT-6 Sol's is twenty times GPT-6 Luna's. Above 272K input tokens OpenAI charges 2x the input and cache rates and 1.5x the output rate "for the full request". The live tables above show SeedRouter's rates for every line.

How is a GPT-6 request billed?

Every response reports its token counts in usage, and the charge follows them:

• input_tokens: the whole prompt, including any cached or cache-written part.
• input_tokens_details.cached_tokens: prompt tokens read from the cache, billed at the cached-input rate.
• input_tokens_details.cache_write_tokens: prompt tokens written to the cache, billed at the cache-write rate.
• output_tokens: the answer plus reasoning, billed at the output rate.

So the cost of one request is:

cost = ((input − cached − cache writes) × input rate
      + cached × cached-input rate
      + cache writes × cache-write rate
      + output × output rate) / 1,000,000

If the prompt is over 272K input tokens, every line uses the long-context rates.

What does a typical request cost?

Take a request with 2,000 input tokens and 1,000 output tokens. Its cost is 2,000 times the input rate plus 1,000 times the output rate, divided by 1,000,000. Because output costs five times input on all three models, the output side of that request costs two and a half times the input side, whichever model you pick. The same request on GPT-6 Sol costs a fifth of GPT-6 Astra and twenty times GPT-6 Luna.

Reasoning changes the output side the most. Higher effort usually spends more reasoning tokens on the same question, so for GPT-6 Sol and GPT-6 Luna try none or low first and raise the effort only where the answers need it.

How do I pay less for GPT-6?

• Pick the smallest model that passes your tests. GPT-6 Luna costs a twentieth of GPT-6 Sol, and GPT-6 Sol a fifth of GPT-6 Astra.
• Lower the effort. Reasoning tokens are output tokens. GPT-6 Sol and GPT-6 Luna accept none, which skips reasoning.
• Reuse your prompt prefix. Cached input costs 10% of the input rate. Keep instructions, tools and reference material at the start and unchanged between requests.
• Stay under 272K input tokens. Above it the whole request, not only the extra part, moves to the higher rate.
• Cap the output. max_output_tokens limits the answer and the reasoning together; the smallest accepted value is 16.

Is GPT-6 free?

Not in the API. On SeedRouter, every new account starts with a small free balance, enough for a few test requests on GPT-6 Sol or many on GPT-6 Luna, and a request that fails is never charged. In ChatGPT, OpenAI says Free and Go users can use GPT-6 Luna in the desktop app; ChatGPT plans are billed separately from the API.

Frequently asked questions

How much does GPT-6 Astra cost?

GPT-6 Astra is the most expensive GPT-6 model: on OpenAI's list its rates are five times GPT-6 Sol's on every line, with output at five times input. The table at the top of this page shows the rate your SeedRouter account pays.

How much does GPT-6 Sol cost?

On OpenAI's list, GPT-6 Sol costs a fifth of GPT-6 Astra and half of GPT-5.6 Sol's promotional price; the table at the top shows the rate you pay. The GPT-6 vs GPT-5.6 guide compares the two generations line by line.

Are reasoning tokens billed?

Yes, as output tokens. usage.output_tokens_details.reasoning_tokens shows how many of the output tokens were reasoning.

Is there a monthly fee?

No. You top up your balance once and spend it per request on any model. The balance does not expire.

Check the cost on your own prompt

Send a prompt in the GPT-6 Sol playground, GPT-6 Luna playground or GPT-6 Astra playground. Each answer shows its token counts and cost, billed exactly like an API call.