Cheapest OpenAI API model: GPT-6 and GPT-5.6 prices compared

Choosing an OpenAI model used to be simple. Today the GPT lineup spans two generations and several tiers, and the price gap between the smallest and largest is enormous. Picking the wrong one can multiply your bill without improving your results, while picking the cheapest one blindly can leave your users with weaker answers. This comparison lays out every current GPT model's API price so you can choose deliberately.

Prices are per million tokens at the time of writing, shown at OpenAI's official list price and at GreenTokens prices.

The GPT lineup at list price

The GPT-6 family comes in three tiers, Luna, Sol and Astra, from lowest to highest price. The GPT-5.6 generation is still available alongside it, with its own Luna, Terra and Sol tiers.

So which is the cheapest OpenAI API model?

By price per token, GPT-6 Luna is the cheapest OpenAI model available: $0.10 per million input tokens and $0.50 per million output tokens. That's less than half the price of GPT-5.6 Luna, and a hundredth of GPT-6 Astra's output price.

A few things stand out from the table. The newest generation is cheaper at the small end, so if you're still calling GPT-5.6 Luna, GPT-6 Luna is worth testing straight away. GPT-6 Sol costs the same per input token as GPT-5.6 Terra but less per output token. And GPT-5.5, the oldest model in the list, is one of the most expensive, which makes it a poor default today.

The cheapest model for each kind of job

The cheapest model overall isn't always the cheapest for your task. A model that needs retries, or produces answers someone has to fix, costs more in the end. As a starting point:

• Classification, tagging, extraction, routing and short rewrites: start with GPT-6 Luna. These tasks are well within a small model's range.
• Everyday assistants, summarisation, customer support and most coding help: GPT-6 Sol is the middle tier, and it's the one to test when Luna isn't enough.
• The hardest reasoning, long multi-step agents and tasks where quality outweighs cost: GPT-6 Astra, used only where the smaller tiers fall short.

A common pattern is to route by difficulty: send every request to Luna first, and escalate to Sol or Astra only for the cases that need it. Even a rough split can cut spend substantially, because most traffic in most products is simple.

The same GPT models at up to 60% off

The cheapest model is one lever; the price you pay for it is another. GreenTokens offers most GPT models well below OpenAI's list price, through the same OpenAI API format your code already uses:

The discount is largest on the bigger models, where it matters most. With 60% off, GPT-6 Sol on GreenTokens costs $0.80 per million input tokens and $4 per million output tokens: a third of what GPT-5.6 Terra costs per output token at list price. GPT-6 Luna is sold at its list price, since it's already priced near cost.

Switching without code changes

GreenTokens serves the OpenAI Chat Completions and Responses APIs, with streaming, tool calling and image inputs, so the official OpenAI SDKs work as they are. In Python or TypeScript you set the base URL to https://api.greentokens.io/v1 and use a GreenTokens key; model names stay the same. Tools such as Codex CLI can use it too, through a custom provider in their config.

Because the same key also reaches Claude, Grok, DeepSeek and other families, you can compare a GPT model against an alternative by changing a single string, without opening a new account.

Cut the bill further

1. Reuse prompt prefixes. OpenAI-format caching is automatic for repeated prefixes, and cached input is billed at a tenth of the input price on GPT-6 models.
2. Limit output with max_tokens and concise instructions. Output tokens cost five to six times input on most GPT models.
3. Give each application its own key with a spend limit, so you can see which feature drives the cost and stop a runaway loop.

Common questions

Is GPT-6 Luna good enough for production?

For focused, well-defined tasks it often is, and at its price it's worth testing first. Run it on a few hundred of your real inputs, compare the results with a larger model, and only move up a tier where the difference matters to your users.

Does a cheaper price mean a different model?

Not with a legitimate provider. You call the same model name, the response returns it, and features like tool calling and image input behave the same. The discount comes from how access is bought, not from changing the model.

The short answer

The cheapest OpenAI API model is GPT-6 Luna, at $0.10 in and $0.50 out per million tokens. The cheapest way to use the more capable GPT models is to route only the hard requests to them, and to pay a discounted price when you do: GPT-6 Sol and Astra are both 60% below list on GreenTokens. Check current prices on our models page before you commit, since both official and discounted prices change over time.