> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rimp.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Multi-model

> How Rimp routes prompts across 8 modalities and ~60 models, and how to compare them side-by-side in one call.

The headline feature: send a single prompt to N models and get all results in one structured response.
No more juggling Replicate, OpenAI, Runway, ElevenLabs, and a dozen other dashboards to find which model handles your use case best.

Rimp spans **8 modalities** behind one API key and one credit balance:

| Modality | Endpoint           | Mode        | What it does                                 |
| -------- | ------------------ | ----------- | -------------------------------------------- |
| Image    | `POST /v1/images`  | sync `200`  | Text-to-image and image-to-image             |
| Video    | `POST /v1/videos`  | async `202` | Text-to-video and image-to-video             |
| Voice    | `POST /v1/voice`   | sync `200`  | Text-to-speech, voice cloning, transcription |
| Music    | `POST /v1/music`   | async `202` | Full songs, with or without vocals           |
| Avatar   | `POST /v1/avatars` | async `202` | Talking avatars and lipsync                  |
| Chat     | `POST /v1/chat`    | sync `200`  | LLM chat completions                         |
| 3D       | `POST /v1/three-d` | async `202` | Text-to-3D and image-to-3D (GLB meshes)      |
| Upscale  | `POST /v1/upscale` | async `202` | Image upscaling and face restoration         |

<Note>
  Costs below are in **credits** — Rimp's universal currency (≈ 1,000 credits per \$1 of underlying
  provider cost). The figures are the **base** cost per unit, exactly as shown in Studio; your
  wallet debit then applies your plan's margin multiplier (Free ×1, Pro & Studio ×1.6, Team ×1.3).
  Call **`GET /v1/models`** for the live catalog. `cr` = credits.
</Note>

Models tagged **Studio+** require a Studio or Team plan. Calling them on Free or Pro returns
`402 plan_upgrade_required`.

## Image models

| Slug                 | Provider          | Tier     | Credits        | Capabilities                                |
| -------------------- | ----------------- | -------- | -------------- | ------------------------------------------- |
| `flux-2-klein-9b`    | Black Forest Labs | fast     | 3 cr / image   | text-to-image                               |
| `seedream-5-lite`    | ByteDance         | fast     | 20 cr / image  | text-to-image, image-to-image               |
| `openai-image`       | OpenAI            | standard | 40 cr / image  | text-to-image, image-to-image               |
| `nano-banana-pro`    | Google (Gemini)   | standard | 40 cr / image  | text-to-image, image-to-image               |
| `qwen-image-2-pro`   | Qwen              | pro      | 40 cr / image  | text-to-image                               |
| `flux-pro`           | Black Forest Labs | pro      | 40 cr / image  | text-to-image                               |
| `flux-pro-i2i`       | Black Forest Labs | pro      | 50 cr / image  | image-to-image                              |
| `imagen-4`           | Google            | pro      | 50 cr / image  | text-to-image                               |
| `recraft-v4-pro`     | Recraft           | pro      | 50 cr / image  | text-to-image (design + SVG)                |
| `flux-2-pro`         | Black Forest Labs | pro      | 60 cr / image  | text-to-image, image-to-image               |
| `ideogram-v2-turbo`  | Ideogram          | fast     | 50 cr / image  | text-to-image (typography)                  |
| `imagen-4-ultra`     | Google            | pro      | 80 cr / image  | text-to-image — **Studio+**                 |
| `ideogram-v2`        | Ideogram          | standard | 80 cr / image  | text-to-image (typography)                  |
| `flux-2-max`         | Black Forest Labs | pro      | 100 cr / image | text-to-image — **Studio+**                 |
| `openai-gpt-image-2` | OpenAI            | pro      | 210 cr / image | text-to-image, image-to-image — **Studio+** |

## Video models

Billed per second unless noted. Image-to-video models accept an `image_url`.

| Slug                | Provider   | Tier     | Credits      | Capabilities                                               |
| ------------------- | ---------- | -------- | ------------ | ---------------------------------------------------------- |
| `runway-gen4-turbo` | Runway     | fast     | 50 cr / sec  | text-to-video, image-to-video                              |
| `kling-v3`          | Kling      | standard | 100 cr / sec | text-to-video, image-to-video                              |
| `seedance-2-0`      | ByteDance  | standard | 120 cr / sec | text-to-video (audio sync)                                 |
| `kling-v3-omni`     | Kling      | pro      | 280 cr / sec | text-to-video, image-to-video (audio + 4K) — **Studio+**   |
| `runway-gen4-5`     | Runway     | pro      | 300 cr / sec | text-to-video, image-to-video — **Studio+**                |
| `openai-sora-2`     | OpenAI     | standard | 300 cr / sec | text-to-video (synced audio) — **Studio+**                 |
| `luma-ray2`         | Luma       | standard | 320 cr / Mpx | text-to-video, image-to-video — **Studio+**                |
| `veo-3-1-fast`      | Google Veo | fast     | 400 cr / sec | text-to-video, image-to-video                              |
| `openai-sora-2-pro` | OpenAI     | pro      | 500 cr / sec | text-to-video, image-to-video (synced audio) — **Studio+** |
| `veo-3-1-standard`  | Google Veo | standard | 750 cr / sec | text-to-video, image-to-video — **Studio+**                |

## Voice models

| Slug                     | Provider   | Tier | Credits          | Capabilities                 |
| ------------------------ | ---------- | ---- | ---------------- | ---------------------------- |
| `inworld-realtime-tts-2` | Inworld    | fast | 20 cr / 1k chars | text-to-voice                |
| `minimax-speech-2-8-hd`  | MiniMax    | pro  | 30 cr / 1k chars | text-to-voice, voice cloning |
| `elevenlabs-scribe-v2`   | ElevenLabs | pro  | 12 cr / min      | transcription                |

## Music models

| Slug                | Provider | Tier     | Credits       | Capabilities                |
| ------------------- | -------- | -------- | ------------- | --------------------------- |
| `minimax-music-2-8` | MiniMax  | standard | 100 cr / song | text-to-music (vocals)      |
| `lyria-3-pro`       | Google   | pro      | 120 cr / song | text-to-music (up to 3 min) |

## Avatar models

Billed per second of output. Two request shapes: text-to-avatar (`text` + optional `avatar_id`/`voice_id`) or lipsync (`video_url` + `audio_url`).

| Slug                 | Provider  | Tier     | Credits      | Capabilities                                |
| -------------------- | --------- | -------- | ------------ | ------------------------------------------- |
| `kling-lip-sync`     | Kling     | standard | 40 cr / sec  | lipsync                                     |
| `sync-lipsync-2-pro` | Sync Labs | pro      | 50 cr / sec  | lipsync                                     |
| `heygen-avatar-v`    | HeyGen    | standard | 67 cr / sec  | lipsync                                     |
| `heygen-avatar-iv`   | HeyGen    | standard | 67 cr / sec  | lipsync — **Studio+**                       |
| `kling-avatar-v2`    | Kling     | pro      | 120 cr / sec | lipsync (realistic + cartoon) — **Studio+** |

## 3D models

Output is a GLB mesh URL. Billed per mesh.

| Slug             | Provider | Tier     | Credits        | Capabilities                                             |
| ---------------- | -------- | -------- | -------------- | -------------------------------------------------------- |
| `trellis-3d`     | TRELLIS  | fast     | 33 cr / model  | image-to-3D                                              |
| `hunyuan-3d-3-1` | Tencent  | standard | 400 cr / model | text-to-3D, image-to-3D                                  |
| `hyper3d-rodin`  | Rodin    | pro      | 400 cr / model | text-to-3D, image-to-3D, rigged characters — **Studio+** |
| `trellis-2-pro`  | TRELLIS  | pro      | 820 cr / model | image-to-3D (PBR materials) — **Studio+**                |

## Upscale models

| Slug                    | Provider    | Tier     | Credits       | Capabilities                              |
| ----------------------- | ----------- | -------- | ------------- | ----------------------------------------- |
| `real-esrgan`           | Real-ESRGAN | fast     | 2 cr / image  | image upscale, face restore               |
| `gfpgan-face-upscale`   | GFPGAN      | standard | 4 cr / image  | face restore                              |
| `recraft-crisp-upscale` | Recraft     | standard | 4 cr / image  | image upscale (commercial-safe)           |
| `topaz-image-upscale`   | Topaz       | pro      | 3 cr / Mpx    | image upscale, face restore — **Studio+** |
| `clarity-upscaler`      | Clarity     | pro      | 17 cr / image | image upscale (Magnific-like)             |

## Chat models

Billed per output token. The figures below are credits per **1M output tokens** — a single
turn typically uses far fewer (the default cap is 1,024 tokens). Charged on `max_tokens`, then
reconciled with actual usage on completion.

| Slug                          | Provider  | Tier     | Credits / 1M tokens | Capabilities                             |
| ----------------------------- | --------- | -------- | ------------------- | ---------------------------------------- |
| `deepseek-v3`                 | DeepSeek  | fast     | 300                 | text                                     |
| `meta-llama-4-scout`          | Meta      | fast     | 350                 | text, multimodal (10M context)           |
| `meta-llama-4-maverick`       | Meta      | standard | 500                 | text, multimodal                         |
| `openai-gpt-4o-mini`          | OpenAI    | fast     | 600                 | text, tool use                           |
| `deepseek-v4-pro`             | DeepSeek  | pro      | 800                 | text, tool use                           |
| `qwen-3-max`                  | Qwen      | pro      | 1,200               | text, tool use                           |
| `anthropic-claude-haiku-4-5`  | Anthropic | fast     | 1,250               | text, tool use                           |
| `openai-gpt-5-mini`           | OpenAI    | fast     | 2,000               | text, tool use, multimodal               |
| `mistral-large-3`             | Mistral   | standard | 3,000               | text, tool use — **Studio+**             |
| `gemini-3-pro`                | Google    | pro      | 7,500               | text, tool use, multimodal — **Studio+** |
| `openai-gpt-4o`               | OpenAI    | pro      | 10,000              | text, tool use, multimodal — **Studio+** |
| `anthropic-claude-sonnet-4-5` | Anthropic | standard | 15,000              | text, tool use, multimodal — **Studio+** |
| `xai-grok-4`                  | xAI       | pro      | 15,000              | text, tool use — **Studio+**             |
| `openai-gpt-5`                | OpenAI    | pro      | 30,000              | text, tool use, multimodal — **Studio+** |
| `anthropic-claude-opus-4-7`   | Anthropic | pro      | 75,000              | text, tool use, multimodal — **Studio+** |

## Comparison API

Send one prompt to several models at once. Works across any models that share a modality.

```ts theme={null}
const cmp = await client.comparisons.create({
  prompt: 'cinematic portrait of a fisherman at golden hour, 35mm film',
  models: ['flux-pro', 'imagen-4', 'ideogram-v2'],
  params_shared: { aspect_ratio: '1:1' },
});
```

### Atomic credit reservation

Rimp sums the estimated cost of all N models and reserves it in **one transaction**. If your wallet can't cover the sum, you get `402 Insufficient credits` and **nothing is charged** — not even the models that would have fit.

### Plan limits

The number of models per comparison is capped by your plan:

| Plan   | Max models per comparison |
| ------ | ------------------------- |
| Free   | 2                         |
| Pro    | 3                         |
| Studio | 6                         |
| Team   | 12                        |

### Polling vs webhooks

A comparison returns `202` with a parent `comparison` object. Two ways to wait for results:

<CodeGroup>
  ```ts SDK helper theme={null}
  const result = await client.comparisons.waitFor(cmp.id);
  // resolves when all child generations reach a terminal state
  result.generations.forEach((g) => console.log(g.model, g.outputs[0]?.url));
  ```

  ```ts Webhook theme={null}
  // Subscribe once:
  await client.webhooks.create({
    url: 'https://app.example/webhooks/rimp',
    events: ['comparison.completed'],
  });
  // Then forget about polling.
  ```
</CodeGroup>

Fetch the full result any time with `GET /v1/comparisons/{id}` — it returns the parent plus
every child generation (status, charged credits, and signed output URLs).

## Capability matrix

Not every model supports every operation. Check the `capabilities` array on each model
(visible in the tables above and in `GET /v1/models`):

```json theme={null}
{
  "slug": "flux-pro-i2i",
  "modality": "image",
  "capabilities": ["image_to_image"]
}
```

Common capabilities: `text_to_image`, `image_to_image`, `text_to_video`, `image_to_video`,
`text_to_voice`, `voice_clone`, `transcription`, `text_to_music`, `lipsync`, `text_to_text`,
`tool_use`, `multimodal_input`, `text_to_3d`, `image_to_3d`, `rigged_3d`, `image_upscale`,
`face_restore`.