Models
Every model you can select, what it costs, and what it can do.
Eleven text models are selectable in Reverie. The model button in the composer shows the one this chat is using and its multiplier; tapping it opens Choose AI Model. Images and video use separate models you don't pick.
How to read the table
| Column | What it means |
|---|---|
| Multiplier | Credits per token. A turn costs total tokens × multiplier, where total tokens is everything sent (character card, persona, world-book entries, chat history, system prompt) plus the reply. |
| Context | How much text the model holds in one request, in tokens. Once a conversation outgrows it, older messages are replaced by summaries. |
| Reasoning | Optional — you can turn thinking on per model. Always on — it can't be turned off. Blank — the model has no reasoning mode. See Reasoning mode. |
| Vision | The model can read images you attach to a message. |
| Tier | Basic works on any account. Advanced requires an active subscription. |
A mature turn on a busy chat runs around 8,000 tokens, so the 0.3× default costs roughly 2,400 credits per reply. Multiply that by the ratio between multipliers to compare models: GLM 5 at 1.3× is about four times the cost of the default for the same turn. How credits are spent
Text models
| Model | Multiplier | Context | Reasoning | Vision | Tier |
|---|---|---|---|---|---|
| Llama 3.1 8B | 0× — free | 131,072 | — | — | Basic |
| DeepSeek V4 Flash | 0.3× | 163,840 | Optional | — | Basic |
| MiMo V2.5 | 0.3× | 262,144 | Optional | — | Basic |
| DeepSeek V3.2 | 0.5× | 163,840 | — | — | Basic |
| GLM 5.3 Flash | 0.5× | 1,048,576 | Always on | Yes | Basic |
| DeepSeek V4.1 Flash | 0.7× | 1,048,576 | Optional | Yes | Basic |
| DeepSeek V4 Pro | 0.7× | 163,840 | Optional | — | Basic |
| DeepSeek R1 | 1.0× | 163,840 | Always on | — | Basic |
| GLM 4.7 | 1.0× | 200,000 | Optional | — | Basic |
| Gemini 3 Flash Preview | 1.2× | 1,000,000 | — | Yes | Advanced |
| GLM 5 | 1.3× | 200,000 | Optional | — | Advanced |
That is the whole list. DeepSeek V4 Flash is the default for new chats, and it is also what Reverie uses behind the scenes for summaries, story turns and the character builder.
Two basic-tier models — GLM 5.3 Flash and DeepSeek V4.1 Flash — carry a million-token context window, so a very large world book or a very long chat does not require a subscription.
Advanced models need a plan. Gemini 3 Flash Preview and GLM 5 are the two advanced models. Without an active subscription the picker shows them locked with This model requires a subscription.
Two models can't use tools
DeepSeek R1 and Llama 3.1 8B have no tool-calling support. In a chat on either of them:
- plugins that need tool calls don't run — the plugin card says when a plugin requires them (Using plugins)
- world-book lore lookup doesn't run, so only always-on entries reach the model (World books)
- the character can't write memories mid-conversation
Story mode only offers models that support tool calls, so neither appears in its model list. Everything else — forking, personas, group chats, voice — works on all eleven.
Where the model is set
The model is stored per character, so switching it in one chat carries to your next chat with that character and leaves every other character alone. Set it from the composer's model button, or in the side panel under Settings → Replies & Generation → AI Model.
If a model is ever retired, chats set to it fall back to DeepSeek V4 Flash. The conversation itself is untouched.
Image models
You don't choose an image model. Reverie routes each request on two things: whether the image is a mature one, and whether you supplied reference images.
| Model | Credits per image | Used for |
|---|---|---|
| Krea 2 Turbo | 10,000 | Default route, no reference image |
| Dark Beast Krea 2 | 15,000 | Mature route, no reference image |
| Krea 2 Identity Edit v1.2 | 30,000 | Default route, up to 2 reference images |
| Dark Beast Krea 2 Identity Edit v1.2 | 30,000 | Mature route, up to 2 reference images |
The mature route needs both your own mature-content setting turned on and a character marked as mature. Image generation needs credits but no subscription. Image generation
Video model
| Model | Credits per clip | Output |
|---|---|---|
| LTX-2.3 Dev Image to Video | 120,000 | One 5-second clip at 24 fps with synchronized audio |
Video requires an active subscription, and only one render runs at a time per account — a second request while one is in flight is refused. The optional step that drafts three motion directions for you is a separate AI call: it comes out of the subscriber free-AI allowance of 30 generations per 5 hours, and costs credits once that allowance is spent. Image to video