AI chat
Talk to leading language models, switch between them and pay per message.
AI chat lets you talk to several language models with the same credit balance. Use it to write texts, answer questions, plan projects or brainstorm prompts for the image studio.
Models and prices
| Model | Credits per message |
|---|---|
| GPT-6 Luna | 1 |
| GPT-5.4 mini | 1 |
| Gemini 3.5 Flash | 2 |
| DeepSeek V4 Flash | 1 |
| Gemini 3.1 Pro | 5 |
| GPT-5.5 | 6 |
| Claude Sonnet 5 | 6 |
| Claude Opus 5 | 16 |
Prices are the minimum per message: long conversations and long answers cost more, and the exact charge is shown in your credit history on the account page. On Standard, Pro and Master plans GPT-6 Luna is unlimited (fair use: up to 300 free messages a day). Start with a 1-credit model for everyday questions. Switch to a larger model for complex writing, analysis or tricky reasoning.
How it works
- Type your message (up to 12,000 characters) and send it.
- The answer streams in as it is written, with Markdown formatting like lists, tables and code blocks.
- You can switch models in the same conversation. The next answer comes from the new model.
- Your chats are saved. On desktop, find them in the sidebar. On mobile, use the Chats… menu.
Long conversations
Only the most recent part of a conversation, about 16,000 characters, is sent to the model. In a long chat, the model "forgets" the oldest messages.
Tip
Start a new chat when you change topic. Answers stay focused, and the model doesn't lose details you care about.
If something important was said long ago, repeat it in your new message.
Good uses
- Ask for image prompt ideas, then paste them into the studio.
- Write product descriptions, captions or emails.
- Plan the shots of a film before building it in Tyzli Movies.
- Summarize or rewrite a text you paste in.
Write 5 detailed image prompts for a cozy autumn coffee shop Instagram post. Include subject, setting, light and style in each.
If something goes wrong
If the model fails to answer, the credits are refunded automatically. If it used its whole reply limit thinking and wrote no text, the tokens it used are charged. See Failed generations and refunds.