Save up to 80% on AI models.

We buy capacity in volume and pass it on with a thin margin, using prompt cache when the model supports it. No magic: wholesale pricing with little fat on top.

It is the same model, with the same weights and the same context window. No hidden quantization, distillation, or a 'similar' model.

No. quebra-galho is 100% prepaid: you add credits via Pix and spend only what you use. No monthly fee, no lock-in, no surprise charge. Without balance, requests stop — we never charge beyond what you deposited.

Yes — and with Cursor, Cline, Aider, Continue, Zed, and any tool that accepts an OpenAI- or Anthropic-compatible API. In most cases you change one environment variable.

Our router fails over automatically: if one route fails, the request takes another in milliseconds. That is how we keep 99.9% uptime — often better than using the provider directly.

The price is set in US dollars and converted to reais at request time, at the rate then in effect. So the amount in reais follows the exchange rate without us having changed the price. The dashboard shows which rate billed each call.

On our side, no: we keep only billing metadata — tokens, model, date — never the text, and none of yours is used for training. What we cannot promise for the provider that processes the request is its retention policy, which is in our Terms.