Models and behavior

The character’s replies are written by an AI model, and the behavior (prompt) sets its manner: how detailed the scenes are, how it keeps the character, what to avoid. The same character sounds different depending on this pair. You can change it at any point in the chat.

🧩 Model and behavior

• Model — the “brain” that generates the replies. Models differ in quality, speed, reply length and price.
• Behavior — instructions for the model, the storytelling style. For example, one behavior makes replies short and lively, another long and literary.
• Preset — a saved “model + behavior” pair together with its generation settings.

🎬 Choosing before the chat starts

You choose the first “model + behavior” pair on the character page, in the “Choose the AI model and its behavior” block. The first card is “Author’s choice”, the second is “We recommend”. If you have chatted with the character before, a model labeled “Last used” appears next to them — the one you chose last time. Under each model is its behavior: click it to choose the pair. The Select AI Model button opens the full list of models. If you have already picked your own model, the button is called “Replace AI Model”.

The model card shows the memory size (for example, 4K or 16K) and the price of a reply: “FREE” or an amount in tokens per message. A plan badge such as Story GOD means the model is available starting from that plan.

🔄 How to change the model in a chat

1. In the chat’s side panel, find the preset block. It has at most two cards: the “model + behavior” pair that is writing the replies right now, and your saved preset marked “Preset”.

2. Click a card — the next replies will be written by that pair.

3. To choose a new pair, click Create preset, pick a model and click Continue. In the second step, choose a behavior and click Save. The behavior card shows its author, how many messages have been written with it, user ratings and the size of the instructions.

Your account has one saved preset. It appears in the side panel of all your chats, and you can pick it in any of them with one click. A new preset replaces the previous one. The “Create preset” button is visible while the block has fewer than two cards. If it isn’t there, choose another pair with the pencil on a card (it appears on hover) or remove the preset with the icon on its card. The trash icon only works in the chat where the preset was saved.

When choosing a model, use the tabs All models, Free, Recommended, Tokens and My plan, the search and the sorting: “Trending”, alphabetical or by price. The model card shows the description, style features (for example, Creative or Immersive), model size, memory size (context) and price. If the model isn’t included in your plan, the card shows the required plan’s badge and an Upgrade plan button. Labels show what CAICHAT recommends (“We recommend”) and what the character’s creator chose (“Author’s choice”).

The creator can lock the model

If you see a lock and the text “The author restricted model selection” where you would choose a model, the character’s creator has disabled changing the model and behavior: the character is designed for exactly this pair.

💰 Free and paid models

• Free models are available to everyone and don’t spend tokens.
• Plan models are included in a subscription: the higher the plan, the more models. See Plans.
• Token models charge tokens for every reply. The price of a reply depends on the model and on the size of the character’s profile — it is shown when you choose a model and in the character’s profile (“Message cost”).

Caution

If the message field shows “The selected AI model is not available with the current subscription” or “You have no tokens for this model”, choose another model or top up your balance.

⚙️ Generation settings

Hover over a preset card in the side panel and click to open Model settings. The pencil next to it lets you choose another model and behavior for this card.

Setting	What it changes
Temperature	Creativity. Closer to 0 (Precise) — more accurate and predictable; closer to 2 (Creative) — bolder and more surprising.
Total Context Tokens	The context: how much of the recent conversation the model reads. The maximum depends on the model and your plan, see Memory.
Max Response Tokens	The maximum reply length.
Reasoning	The model first “thinks over” the reply. Available only for paid models and uses part of the reply limit.

The Advanced Settings section has parameters for experienced users:

• Top-P and Top-K — how wide a choice of words the model allows. Lower means stricter replies, higher means more varied ones.
• Presence Penalty — a penalty for repeating topics already mentioned: the higher it is, the more readily the model brings in something new.
• Frequency Penalty — a penalty for repeating the same words often.

Click Apply Settings to save the changes. Not all models support all parameters: unsupported ones are simply not shown.

Tip

If the character starts repeating itself or replies in a formulaic way, try raising Temperature and Presence Penalty a little. If it starts talking nonsense, lower Temperature.

🧠 Reasoning

Some models can “think” before replying: this improves the logic of complex scenes, but the reply arrives more slowly. Reasoning is turned on in the preset settings. For some models it is always on and can’t be turned off.

If the model has spent the whole reply limit on thinking, the chat shows “The model reached the response limit and returned only thoughts”. Increase Max Response Tokens in the settings and regenerate the reply.

📚 Models and behaviors catalog

In the left menu, in the Studio group, there are Models and Prompts sections — they collect all the models and behaviors on the platform: descriptions, the best characters on each model, creators’ recommendations. Creators also publish their own behaviors (prompts) there, and owners of their own AI models can connect them via API.

❓ Related questions

Which model should I choose?

Start with the models labeled “We recommend” or the one the character’s creator chose. If the replies seem weak, try a model one level higher: the difference in quality between models is noticeable.

How much does one message cost?

It depends on the model and on the size of the character’s profile. The price is shown when you choose a model, and in the character’s profile in the “Message cost” field. On models included in your plan, messages don’t spend tokens.

The model changed by itself

If the selected model became unavailable (for example, your subscription or tokens ran out), a hint appears in the message field. Choose another model in the chat’s side panel.