OpenAI API pricing in 2026—how much GPT costs per request, and how to pay with Pix

The OpenAI API charges per token, not per month. Input and output have separate prices, and output is the more expensive part. On Kunavo, the range goes from $0,04 / $0,20 per 1 million tokens on GPT-6 Luna to $4,00 / $20,00 on GPT-6 Astra, OpenAI’s current top-tier model; between them are GPT-6 Sol, at $0,80 / $4,00, and the workhorse GPT-5.6 Terra, at $0,70 / $4,20 — 60% to 65% below OpenAI's list price, depending on the model. A ChatGPT subscription is a separate matter: it does not include API usage, and the API does not require a subscription. From Brazil, you can pay with Pix.

Prices checked on October 3, 2026. This page reads Kunavo prices directly from the model catalog — what appears here is what the API charges. The “OpenAI list price” column is the price published by OpenAI for the same model at openai.com/api/pricing. Kunavo is a gateway that resells access to models, not an independent analysis site.

ChatGPT Plus and the OpenAI API: two products, two charges

People searching for “how much does the ChatGPT API cost” almost always want to know something else first: whether the subscription they already pay for covers the API. It does not.

The subscription does not provide an API key or API credit, and API credit does not increase ChatGPT limits. Plan prices are at openai.com/chatgpt/pricing; they change frequently, which is why they are not listed here. This page covers only the API — the part that costs per token.

OpenAI API pricing by model

In dollars per 1 million tokens, as Kunavo charges. The “OpenAI list price” column is the published price for the same model.

Live prices are always on the pricing page and on each model page, such as GPT-6 Sol or GPT-6 Luna. Which one to use for what:

• gpt-6-astra — OpenAI’s top-tier model: cutting-edge reasoning and long agent sessions. For the minority of difficult tasks.
• gpt-6-sol — OpenAI positions it for complex code and agent workflows. On Kunavo, $0,80 / $4,00: below GPT-5.6 Terra on output, slightly higher on input. The sensible default for code and agents.
• gpt-6-luna — according to OpenAI, the most efficient model for well-defined, high-volume tasks such as classification and extraction. Kunavo’s least expensive GPT.
• gpt-5-6-terra — the workhorse of the 5.6 family: RAG, assistants, and most production tasks.
• gpt-5-6-sol and gpt-5-5 — previous-generation top-tier models. At Kunavo, both cost the same per token; anyone still running gpt-5-5 can switch to the newer generation by changing only the model name.

GPT-6 Sol and GPT-6 Luna have the same context window, 1.050.000 tokens, and return up to 128.000 output tokens. An additional charge for very long requests: above 272,000 prompt tokens, GPT-5.5, the GPT-5.6 family, and all three GPT-6 models — Astra, Sol, and Luna — charge the entire request at 2× the input price and 1.5× the output price. Anyone sending an entire codebase in one call should know this limit: it matters more than the model choice.

What a real request costs

At Kunavo prices, without caching and reasoning tokens — in other words, the floor of what appears on the bill:

The minimum $10 top-up yields approximately 1.590 RAG responses on GPT-5.6 Terra, or approximately 29.400 on GPT-6 Luna. Between GPT-6 Luna and GPT-6 Astra, the output price difference is approximately 100 times — for a simple task, model choice is the biggest cost lever available.

Two pitfalls every estimate gets wrong

Reasoning tokens. Models such as GPT-6 Astra and GPT-6 Sol think in internal tokens before responding. They do not appear in the text, but they are charged at the output price. The examples above do not include them, so the actual bill is higher for a heavy task. The exact number is included in each response, in usage.

Retries. A response that returned successfully and was discarded by your code — wrong format, another attempt — has already been paid for. In agents, one visible task commonly consists of a dozen billable calls. Requests that fail within Kunavo are not charged.

Prompt cache. If your app always sends the same long system prompt or the same documents, cache hits cost a fraction of the input price — on GPT-5.6 Sol, $0,20 instead of $2,00 per 1 million tokens. How to achieve cache hits is covered in the cache documentation.

Paying for the OpenAI API from Brazil without an international card

Kunavo charges through Stripe, using a prepaid dollar balance, with no subscription and no OpenAI platform account. At checkout, people paying from Brazil see:

• Pix — the amount appears in reais; you scan the QR code or use copy and paste, and the balance is credited immediately
• Visa, Mastercard, American Express, JCB, and UnionPay cards — an international card incurs IOF, as with any purchase abroad
• Apple Pay, Google Pay, and Link

The minimum top-up is $10; larger top-ups include a bonus: $100 viram $110,00, $1.000 viram $1.200,00, $5.000 viram $6.250,00. The balance never expires and failed requests are not charged. The step-by-step Pix guide, from sign-up to the first call, is at how to pay for an AI API with Pix.

Switching existing OpenAI code

Kunavo is OpenAI-compatible: change the base_url and the key, and nothing else. chat.completions and responses retain the same format, and the response includes the usual usage field, which lets you verify every charge.

from openai import OpenAI

client = OpenAI(
    api_key="sk-kn-...",
    base_url="https://api.kunavo.com/v1",  # no lugar de api.openai.com
)

resp = client.chat.completions.create(
    model="gpt-5-6-terra",
    messages=[{"role": "user", "content": "Resuma esta nota fiscal em três frases."}],
)
print(resp.choices[0].message.content)
print(resp.usage)  # tokens de entrada, saída e raciocínio: a base da cobrança

The key (sk-kn-…) is created in the dashboard after sign-up; you can also set a monthly spending limit per key there. The steps are in the quickstart.

When going directly through OpenAI is better

Kunavo is a gateway with shared capacity: there is no quota dedicated to your account and no contractual SLA. Anyone who needs a guaranteed quota or an SLA — a product with availability promised to the customer — should buy directly from OpenAI.

And for people who spend the entire day in the chat window, a ChatGPT subscription is usually less expensive. The calculation is a division: plan price ÷ cost of one chat round. A round of 2,000 input tokens and 700 output tokens costs approximately $0,0124 on GPT-5.6 Sol and $0,00434 on GPT-5.6 Terra. Anyone who makes more rounds month after month than that division yields — within the plan’s limits — comes out ahead with the subscription. The API wins for software, automation, and variable usage, where a month with no usage should cost nothing.

To recalculate the figures with your numbers, use the OpenAI API pricing calculator. The Anthropic equivalent is at Claude API pricing.

Frequently asked questions