Help

Free models cost nothing; they are limited by number of calls, as “What are the limits on free models?” explains. Drawing is free too, up to 10 pictures a day per account. Claude, GPT and other big models and the DeepSeek coding line are paid: you top up a balance in dollars and they are charged to it by usage, at the prices on the Pricing page. Free models run on the free allowances their providers offer, which the site brings together in one place; big models are paid for by the site, and top-ups cover that cost. Prompts are not sold. The full picture is on the about page.

Free models are not charged; they are limited by number of calls: 30 calls per account every 5 hours and 60 per network. When the calls shared by the whole site run out, free models pause for everyone for a while. A longer message counts as several calls. Its length is counted over everything sent with it: the message together with the conversation so far, the persona and any attached file text. Each Chinese character counts 1 and every 4 other characters (letters, digits, spaces, punctuation) count 1, and each 50,000 counted, or part of it, is one call. So 200,000 Chinese characters count as 4 calls, while the same length of English text or code counts as about 1; once a conversation has reached 100,000 counted, even a short new message counts as at least 3. If fewer calls are left than a message needs, it is not sent, and the page says how many calls it needs, how many are left in this window and when more come back. Claude, GPT and other big models are charged by usage to your balance and do not count toward these calls.

Two reasons, both practical. A balance stored only in a browser vanishes the moment site data is cleared; tied to an account, it is the same on every device. And the free calls, the daily pictures and the free tries are counted per account, which only works when the account is the same on every visit. Signing in is free and takes one click with Google or GitHub, or a code sent to your email.

Nothing is lost. Sign in again and the My account page issues a new key on the same account, with the same balance. If you think a key has leaked, revoke it from the My account page — the balance is unaffected.

The amount you pay becomes your balance, shown in dollars. Claude, GPT and other big models and the DeepSeek coding line are charged to it by usage. Free models do not touch the balance; they are limited by calls. The balance never expires and nothing renews automatically; top up again whenever it runs out.

Yes, up to 10 per account — useful for separating projects. All keys on an account draw from the same balance and share the account’s call limits, so more keys do not mean more calls. Opening extra accounts to get around those limits is not allowed, and the number of new accounts per network is capped each day.

On the chat page, yes: conversations are saved to your account so you can return to them, up to 50, and you can delete any or all of them on the chat page. For requests sent directly to the API, an excerpt of each request and reply is kept (the latest 100 per account, for at most 30 days), and the time, the model chosen and the amount used are recorded, for billing and for counting calls. The operator can read chats and API call content in the administration area, to understand how the site is used. Either way, messages are forwarded to a third-party model provider, and some providers, particularly on free tiers, may retain what they receive or use it to improve their services. We do not use your conversations to train models. Avoid sending sensitive personal information. Details are in the privacy policy.

Not advisable. This is a free beta with no uptime promise and no support commitment. It is built for trying things out, prototyping, and side projects. If something depends on being available at 3am, buy an API key from a provider who will answer the phone.

The current list is on the models page, and the My account page shows what your key can call right now. The list can change: models get added, and a model can disappear when the upstream provider retires it.

Yes, to the models marked with the image sign in the model lists. On the chat page, add up to 4 images to a message with the image button, or paste or drop them into the box. On big models each image adds a fixed charge the first time it is sent in a conversation, and the estimate beside Send includes it; on free models images cost nothing. Images are passed to the model provider for the answer and are never stored: after a reload, the conversation shows that a message had images, not the images themselves. Through the API, see the images section of the docs.

Yes. On the chat page, use the button beside the message box or drop the file into it: PDF, TXT, Markdown, CSV, JSON and common code files, up to 4 per message and 20 MB each. The page reads the text out of the file in your browser and sends that text with your question (up to 40,000 characters), so any model can take it; it counts toward the length of the message: on big models it is charged like the rest of the message, and the estimate beside Send includes it; on free models it is not charged, and a long message counts as several calls. The text stays with the conversation, so you can keep asking about the file after a reload. A scanned PDF with no text layer cannot be read; Word documents are not taken yet.

Almost always a max_tokens value set too low. These models produce reasoning before the visible answer, and if the budget runs out during that stage the visible part never gets written. Raise it and try again.

The balance pays for Claude, GPT and other big models and the DeepSeek coding line; free models do not touch it, so you can keep using them with an empty balance. Before you send, the line beside Send says what would cover the turn: a free model, or a fresh conversation, which costs less. To keep using the big models, top up on the Pricing page.

Sign in, open the Pricing page and select Top up; the smallest top-up is $5. The amount you pay becomes your balance, shown in dollars. It pays for the DeepSeek coding line, which is not limited by the free-model call counts, and for Claude, GPT and other big models, all charged by usage; free models do not touch the balance. The balance never expires and nothing renews automatically; top up again whenever it runs out. For 35 days after a payment, free models allow 90 calls every 5 hours. If you pay with an email other than the one you sign in with here, a claim code is sent to that email; enter it on the My account page. Claude, GPT and other big models are open to accounts that have topped up.

An unused balance can be refunded; the part already used cannot. To ask for a refund, write to us through the contact page with the email address used for the payment.

No. The balance cannot be transferred to another account or withdrawn as cash. Sharing a key so several people can spend one balance is also not allowed: a key seen in use from several places at once gets disabled.

The site no longer uses credits. Free models are limited by calls and the big models are charged in dollars; credits given out earlier are no longer in use.