AI chat: how to use it

Which model to pick

Models differ in smarts, speed and price. The comparison with prices is on the chat models page; you can switch the model at any time with the Model list above the input field.

• Claude Opus — the hardest tasks, code and long texts. The most expensive.
• Claude Sonnet — smart and fast: writing, code, analysis.
• GPT‑6 Sol — a strong all-round model.
• GPT‑6 Luna and Gemini Flash — fast and inexpensive, for everyday questions.
• Grok — reasoning and code.
• DeepSeek — fast and cheap, but text only: photos, PDFs and drawing are unavailable; you can attach text files only.

Photos and documents

Attach files to a message with the paperclip button next to the input field. What you can attach depends on the selected model.

• Images (JPEG, PNG, WebP, GIF) — photos, screenshots, diagrams; for every model except DeepSeek. The model describes them, finds a mistake, reads the text.
• PDF — for Claude, GPT‑6 and Gemini; Grok and DeepSeek do not accept PDFs. Text files (TXT, Markdown, CSV, JSON) are accepted by every model. The model summarises, answers questions about the content, compares.
• Each model has its own limit: up to 5 files per message, each up to 5 MB for Claude and up to 10 MB for the others; DeepSeek takes up to 3 files of 2 MB.
• If the model does not accept a file of that type or size, the studio tells you before sending.

Images right in the conversation

Ask the model to draw: “draw”, “make a cover”, “show me what it could look like”. It writes a detailed prompt itself and hands it to an image model. When the image is ready, it appears in the conversation, where you can open and download it.

• The image model is chosen with the “Model for images” list above the input field (shown for models that can draw); the default is Nano Banana 2.
• The image is billed at the price of the selected image model (shown in the list), separately from the answer.
• Models with tool support can draw — DeepSeek cannot.

Chats and history

• Every conversation is a separate chat in the list on the left, grouped by day: today, yesterday, earlier.
• Rename or delete a chat; a deleted chat cannot be restored.
• The model remembers the whole conversation in the current chat. For a new topic, start a new chat — the model will not get confused and answers will cost less.
• If an answer was cut off, ask the model to continue.
• When a conversation gets too long for the model, the studio suggests starting a new chat.

What an answer costs

An answer is billed by what it actually used: how much text the model read — your message, the chat history and attachments — and how much it wrote. The amount is shown under every answer. The model picker shows the approximate price of a typical answer.

• A long history and large files make every next answer more expensive: the model reads them again.
• For simple questions pick a fast, inexpensive model — it costs noticeably less than a top one.
• If you run out of tokens, the studio offers to top up your balance.

How to ask a question in the chat

1. Open the chat. In the web studio, open Chat and press New chat.
2. Pick a model. Choose a language model in the Model field; the approximate price of an answer is shown next to it.
3. Write a message. Describe the task and, if needed, attach a photo or a document with the paperclip button.
4. Ask for an image. To get an image, just say so: “draw…”. The image model is chosen with the list above the input field.
5. Keep talking. Refine and edit — the model remembers this chat’s conversation.