> For the complete documentation index, see [llms.txt](https://docs.userouter.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.userouter.ai/studio/chat.md).

# Chat

Pick a model, write a message, and see the exact cost of every reply.

## Start a chat

1. Open **Studio → Chat**.
2. Pick a model. The default is **Claude Sonnet 5.5**.
3. Type your message (up to **40,000 characters**) and send.

The reply appears as it's written and is shown as formatted text with a **Copy** button.

## Models

| Model             | Made by   | Good for                     |
| ----------------- | --------- | ---------------------------- |
| Claude Sonnet 5.5 | Anthropic | Default. Strong all-rounder. |
| Claude Opus 5.5   | Anthropic | Premium. Hardest tasks.      |
| GPT-6 Sol         | OpenAI    | Balanced.                    |
| Gemini 3.8 Flash  | Google    | Fast and cheap.              |
| DeepSeek V4 Pro   | DeepSeek  | Budget.                      |

Prices: [Model catalogue](/get-started/models.md#chat).

## What a reply costs

Chat is priced per token, with separate input and output rates.

1. **When you send**, Router reserves enough credits for your message plus the longest possible reply (**4,096 output tokens**).
2. **When the reply finishes**, you're charged for the tokens actually used. The rest of the reservation is refunded straight away.

Each reply shows its credits, the dollar amount, and the input and output tokens. If a model doesn't report token counts, Router estimates them and marks the reply **(estimated)**.

Because the reservation covers the longest possible reply, your balance briefly dips by more than the final charge. You need enough credits for the reservation to send a message.

## Conversations

* **New chat** starts a fresh conversation. Past conversations are in the list on the side.
* Router remembers context automatically. It includes the most recent **48** turns of the conversation, up to about **128 KB** of text. Older turns drop out of the model's context but stay in your history.
* Replies are capped at **4,096 output tokens**.
* You can run one chat reply at a time. Image and video requests can run alongside it.