Claude Haiku 4.5 per token, in reais

Frontier O mais rápido e barato da linha Claude. Indicado para classificação, extração e alto volume.

Where it does well

Classification, extraction and short answers at volume, with low latency. Context window of 200,000 tokens.

Where it is not the best choice

Long reasoning tasks: it answers fast, not deep.

What changes when you pay in reais

Buying direct from the maker means paying in dollars plus two things that are not on their price list: Brazil's 3.5% IOF tax and your card issuer's FX spread, usually another 2% to 4% and never itemised. Here you are billed in reais, with neither.

The price in reais is also fixed: when the dollar rises, the maker price rises on your statement and this one does not.

How to use it on GPUBrasil

The API speaks the OpenAI format. Swap the base_url, use a gpub_live_ key and pass claude-haiku-4-5 as the model.

from openai import OpenAI

client = OpenAI(
    base_url="https://gpubrazil.com/v1",
    api_key="gpub_live_...",
)

r = client.chat.completions.create(
    model="claude-haiku-4-5",
    messages=[{"role": "user", "content": "Explain what a token is."}],
)
print(r.choices[0].message.content)

The same endpoint works in Cursor, Cline, Continue, Zed, Aider, Open WebUI and LibreChat — they all accept an OpenAI-compatible endpoint.

Create an account →   See every model and price →

FAQ

How much does Claude Haiku 4.5 cost on GPUBrasil?

R$ 12.95 per 1 million input tokens and R$ 60.90 per 1 million output tokens, billed in Brazilian reais. No subscription and no minimum purchase — you pay only for the tokens you use.

How do I call Claude Haiku 4.5?

The API is OpenAI-compatible. Point base_url at https://gpubrazil.com/v1, use a gpub_live_ key and pass claude-haiku-4-5 as the model. It works with the OpenAI SDK and with tools like Cursor, Cline, Continue, Zed and Open WebUI with no adaptation.

What is the context window of Claude Haiku 4.5?

200,000 tokens, counting what you send plus what the model answers in the same call.

Similar models