GPT API pricing: Astra, Sol, Terra and Luna per 1M tokens

OpenAI's newest GPT models are priced per million tokens, and on all four of them output costs five times input. Luna is the cheapest, Terra costs ten times Luna, Sol twice Terra, and Astra five times Sol. Caching, batch and long prompts then change the bill. Below: the official rates, how a bill adds up, and our prices.

What does the GPT API cost per million tokens?#

OpenAI prices each model per million tokens, with separate rates for input, cached input and output. These are the Standard list prices for the four newest models, Astra, Sol, Terra and Luna, with the modifiers that apply on top. Sol has two current versions, 6.0 and 6.1, at the same input and output price.

Three things on that page are easy to miss. First, cached input usually costs a tenth of the input price, but on Sol 6.1 and Terra it costs a twentieth. That makes Sol 6.1 the better buy of the two Sols: the token prices are equal and its cache costs half as much. Second, a request with more than 272K input tokens is billed at long-context rates: twice the input price and one and a half times the output price, and the cache prices double too. Third, on these models writing to the cache costs extra: 1.25 times the input price for the default 30-minute cache, or 2.5 times for the optional 12-hour cache.

How is a GPT API bill calculated?#

Each request is the sum of up to five parts, each with its own rate:

• Input tokens: everything you send that is not read from the cache, including the instructions, the conversation so far and the tool definitions.
• Cached input tokens: when a request starts with the same prefix as a recent one, that part is billed at the cached-input rate. Caching is on by default for prompts of 1,024 tokens or more. The response shows the count in usage.input_tokens_details.cached_tokens in the Responses API, or in usage.prompt_tokens_details.cached_tokens in Chat Completions.
• Cache writes: storing a new prefix costs 1.25 times the input rate for 30 minutes, or 2.5 times for 12 hours.
• Output tokens: everything the model writes, including its hidden reasoning. In OpenAI's Responses API, the field usage.output_tokens_details.reasoning_tokens shows how many of the billed output tokens were reasoning.
• Tools: web search is charged per call, plus the search content tokens it adds, billed at the model's rates.

Then the processing tier changes the price of every token. Batch and Flex cost half the Standard price for work that can wait, and a batch finishes within 24 hours. Fast mode, called Priority processing until July 30, 2026, costs twice the Standard price. Ultrafast costs six times as much and is offered only on Astra and Sol 6.1. Regional processing (data residency) adds 10% on models released on or after March 5, 2026.

Reasoning depth is the other lever you control. A lower reasoning effort usually means fewer reasoning tokens billed as output, at some cost in quality on hard tasks. Our Responses reference lists the effort values it accepts.

A worked caching example#

Say an agent sends the same 100,000-token set of instructions ten times within 30 minutes on Sol 6.1. Without caching, you pay for 1,000,000 input tokens at the full input rate. With caching, you pay for one write at 1.25 times the input rate and nine cached reads at a twentieth of it. In input-rate units that is 125,000 plus 45,000, or 170,000 instead of 1,000,000: about 83% less on that part of the bill. On Sol 6.0, where cached input costs a tenth, the reads come to 90,000 and the saving is about 79%. Output does not change, and a cache hit is likely but not certain, so check the cached-token field on real traffic. Prompt caching explains how caching works with GPT models here.

Is the newest generation cheaper than the last one?#

Mostly, yes. At the prices OpenAI lists today:

• Sol 6.1 costs half as much as gpt-5.6-sol on input and output, and a quarter as much on cached input. OpenAI notes that the current price of the older Sol is promotional and available at least through November 21, 2026.
• Terra costs half as much as gpt-5.6-terra on input, less than half on output, and a quarter on cached input.
• Luna costs half as much as gpt-5.6-luna on input and cached input, and less than half on output.
• Astra is the exception. It is OpenAI's most capable model, and it costs twice as much as gpt-5.5 on input and two-thirds more on output.

Moving to the newer generation cuts the token price, but test quality on your own tasks before you switch a workload.

What do the same models cost through 5abeer?#

5abeer sells GPT models through an OpenAI-compatible API, paid from your balance, below the official list price. The list below reads our current prices from the live catalog, so it stays correct when they change.

The OpenAI models we offer on the API today:

Every model, with its prices, is on the models page, and each model's own page lists our cache prices. The caching guide shows where each response reports cached tokens.

Which GPT model should you pick?#

Price per token is only half the decision. A cheaper model that needs three attempts can cost more than a stronger one that gets it right the first time. OpenAI's own advice is to start with Astra if you are not sure. From there:

• Astra is OpenAI's most capable model, for the most demanding work. At five times Sol's token price, use it where you have measured a quality gap.
• Sol 6.1 comes close to Astra on complex work at a fifth of the price. Prefer it to Sol 6.0: the token prices are the same and its cache costs half as much.
• Terra is OpenAI's fast, efficient model for visual reasoning, computer use and clear writing, at half of Sol's price.
• Luna is the cheapest, built for focused, high-volume tasks such as short classification or extraction jobs.

A common pattern is to split work by task: the cheap model for the bulk of simple requests, and the strong one for the steps that need it. Because every model is billed per token from the same balance, switching costs nothing beyond the tokens. For coding agents, see Codex with 5abeer, the Codex setup docs and how to run Codex CLI with an API key.

How do you buy GPT API access through 5abeer?#

1. Create a key in your dashboard. API keys explains the spend limit, expiry date and IP allow-list you can set on each key.
2. Add balance. API usage is deducted from your balance, token by token.
3. Point your client at our OpenAI-compatible base URL and use a model id from the list above. The OpenAI API page lists the base URL and current prices, and Base URLs gives the address for each API format.

Batch, Flex, Fast and Ultrafast are OpenAI's processing tiers. Our docs do not list them, so plan your costs with us on regular requests. To compare GPT with the Claude and Gemini models on the same balance, see the API page. For the app plans, see the apps and pricing.

Questions#

Are reasoning tokens billed as output?#

Yes, even though the response does not show them. In OpenAI's Responses API, the usage field output_tokens_details.reasoning_tokens counts them, and a lower reasoning effort usually means fewer.

Does a long prompt cost more per token?#

Yes, past 272K input tokens. Such a request pays twice the input price, one and a half times the output price and double the cache prices. Shorter requests pay the Standard rates.

Is caching automatic?#

Yes, for prompts of 1,024 tokens or more. The request that stores a prefix pays 1.25 times the input price, so caching pays off only when the prefix is read again before it expires. Put the stable part of the prompt first to raise the hit rate.

Is the Batch API worth it?#

If your requests can wait up to 24 hours, it halves the input and output prices at OpenAI. Flex processing gives the same discount on regular requests that may run more slowly. Our docs do not list either option.

Do Responses and Chat Completions cost the same?#

Yes. OpenAI does not price the Responses, Chat Completions, Realtime, Batch or Assistants APIs separately: you pay the model's token rates. Both are available with us, in the Responses and Chat Completions references.

How is web search charged?#

Per call, at the price per 1,000 calls in the table above, plus the search content tokens it adds, billed at the model's rates.

Is a ChatGPT subscription cheaper than the API?#

It depends on how much you use and where. A subscription works inside the apps, while the API is paid per token from your balance. AI subscription vs API explains the difference and which one you need.

Sources#

Checked on 2026-11-01:

• Pricing, OpenAI API docs
• Models, OpenAI API docs
• Prompt caching, OpenAI API docs
• Reasoning models, OpenAI API docs
• Batch API, OpenAI API docs