Pertanyaan yang sering diajukan

Tidak. Kalau permintaan gagal, baik kami yang menolaknya atau penyedia model yang mengembalikan error, tidak ada biaya. Kamu hanya dikenai biaya saat model mengembalikan respons.

Satu pengecualian adalah respons streaming yang berhenti dibaca klienmu di tengah jalan, misalnya karena batas waktu atau pengguna membatalkannya. Generasi langsung berhenti, dan kamu membayar input serta output yang sudah dihasilkan sampai titik itu.

Credit usually shows up within seconds of payment. If it doesn’t, don’t pay again. Completed payments are credited automatically, and if something goes wrong on our side, we retry every few minutes. Refresh the Wallet page to check.

If it still isn’t there after two hours, email support@cheapaimodel.ai from the address you sign in with, and include the receipt Stripe emailed you or the time you paid.

Many apps send extra requests in the background with your key. Open WebUI, for example, asks a model to write a title and tags for each chat, and Aider sends extra requests to write commit messages and summarize long chats. Each one is a separate request, billed on its own, and some apps use a different model for these tasks.

To see exactly what was sent, filter the Logs page by key and model. If you don’t need these extras, turn them off in the app or point them at a cheaper model.

Models don’t remember earlier messages, so chat apps resend the whole conversation with every new message. Your input tokens, and your cost, grow with each turn. Coding agents like Cline and Aider send even more, including file contents and tool output.

To keep costs down, start a new chat when you switch topics, or limit how many past messages the app sends. If a model’s page lists a cache read price, repeated input that hits the cache is billed at that lower price.

Not right now. Per-token prices are the same no matter how much you add. Your savings come from per-model discounts, which apply automatically to every request; see Pricing. If we add top-up bonuses later, the Wallet page will show the bonus before you pay.

Your request went to OpenAI, not to us. You set your cheapaimodel key but not the base URL, so the SDK or tool fell back to its default, https://api.openai.com/v1, and OpenAI doesn’t recognize the key.

Set the base URL to https://api.cheapaimodel.ai/v1 as well. With the OpenAI SDKs, you can also set the OPENAI_BASE_URL environment variable. Base URL lists the right setting for each SDK and tool.

Asking a model what it is won’t tell you much. Models often get their own name, version or knowledge cutoff wrong, and official apps fix that with a system prompt the API doesn’t include. Your request goes to the model named in its model field, and the Logs page shows which model each request was billed as.

Answers can also differ from the official app because the app may search the web and uses its own default settings, such as the reasoning mode. For a fair comparison, call the provider’s own API with the same parameters.

Up to 60 minutes. A streamed response is also closed if no data arrives for 5 minutes.

Your client may give up sooner. The OpenAI SDKs time out after 10 minutes by default, so for long reasoning or long outputs, raise the client timeout or use streaming. A request that times out can be retried; see Errors for which errors are safe to retry.

The API reference lists every endpoint the gateway serves, and each model works only on the endpoints shown on its page in the model catalog. Fine-tuning, the Files, Batch and Assistants APIs, and the image and audio APIs aren’t available yet.