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

# Providers

> The AI models, voice engines, and transcribers available in Orato, and how to choose the right combination.

Every assistant combines three providers: a **transcriber** (ears), an **AI model** (brain), and a **voice engine** (mouth). The combination affects how fast the assistant responds, how natural it sounds, which languages it handles, and what it costs per minute.

> [!NOTE]
>
> Model lists reflect the dashboard at the time of writing. The assistant editor always shows what's currently available, along with a live cost estimate.

## AI models

Set on the [Settings](/assistants/settings#ai-brain) tab.

**Anthropic**

| Model | Notes |
| - | - |
| **Claude Haiku 4.5** | Default. Fast and affordable, strong at following instructions. |
| **Claude Sonnet 4.6** | More capable reasoning for complex sales and support |
| **Claude Opus 4.7** | Most capable, for demanding or sensitive conversations |

**Groq**

| Model | Notes |
| - | - |
| **GPT-OSS 120B** | Open-weight model on Groq's fast inference, with a very low LLM cost per minute |

**OpenAI**

| Model | Notes |
| - | - |
| **GPT-4o** | Premium. Well-rounded quality. |
| **GPT-4.1** | Premium. Strong instruction following. |
| **GPT-4o Mini** | Premium. Lighter and faster. |

**Sarvam AI**

| Model | Notes |
| - | - |
| **Sarvam 105B** | Built for Indian languages |

**Google**

| Model | Notes |
| - | - |
| **Gemini 2.5 Flash-Lite** | Lightweight and fast |
| **Gemini 2.5 Flash** | Balanced speed and quality |
| **Gemini 3.5 Flash-Lite** | Newer lightweight model |
| **Gemini 3.5 Flash** | Newer balanced model |

## Voice engines

Set on the [Voice](/assistants/voice-and-language) tab.

| Provider | Example models | Speed range | Strengths |
| - | - | - | - |
| **Cartesia** | Sonic 3 | 0.6x to 1.5x | Natural, low-latency voices, including Indian voices. Default. |
| **ElevenLabs** | Turbo v2.5, Flash 2.5 | 0.7x to 1.2x | Very expressive voices |
| **Sarvam** | Bulbul v3 | 0.5x to 2x | Hindi and Indian regional languages |
| **Smallest AI** | Lightning v3.1 Pro | 0.5x to 2x | Lowest voice cost per minute |

## Transcribers

Picked automatically from your languages. Override on the [Advanced](/assistants/advanced#transcriber) tab.

| Provider | Models | Use for |
| - | - | - |
| **Deepgram** | Flux English (Recommended), Flux Multilingual | English, or callers who mix languages |
| **Sarvam AI** | Saaras v4 Realtime, Saaras v3 | Hindi and Indian regional languages |

## Recommended combinations

| Use case | AI model | Voice | Transcriber |
| - | - | - | - |
| English sales or support (US or India) | Claude Haiku 4.5 | Cartesia Sonic 3 | Automatic |
| Complex, high-value conversations | Claude Sonnet 4.6 | ElevenLabs | Automatic |
| Hindi or regional-language calls | Sarvam 105B or Claude Haiku 4.5 | Sarvam Bulbul v3 | Sarvam AI |
| High-volume, cost-sensitive campaigns | GPT-OSS 120B (Groq) or Gemini Flash-Lite | Smallest AI | Automatic |

> [!TIP]
>
> Always test a new combination with **Talk** before a campaign. Latency and voice quality are easier to judge by ear than on paper.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.