Documentation 2026-07-28 15:58 UTC to 2026-07-29 17:58 UTC

1#### Model Capabilities1#### Model Capabilities

2 2 

3# Speech to Speech API3# Speech to Speech

4 4 

5Build real-time voice applications powered by Grok. Stream audio and text bidirectionally via WebSocket for voice assistants, phone agents, and interactive voice systems.5Build real-time voice applications powered by Grok. Stream audio and text bidirectionally via WebSocket for voice assistants, phone agents, and interactive voice systems.

6 6 

133 133 

134| Model | Description |134| Model | Description |

135|-------|-------------|135|-------|-------------|

136| `grok-voice-latest` | Alias that tracks the newest voice model |136| `grok-voice-latest` | Alias for `grok-voice-think-fast-1.0`.Updates to `grok-voice-think-fast-2.0` on August 5, 2026. |

137| `grok-voice-think-fast-1.0` | Flagship voice model |137| `grok-voice-think-fast-2.0` | Flagship voice model |

138| `grok-voice-think-fast-1.0` | Previous-generation voice model |

138 139 

139## Session Parameters140## Session Parameters

140 141 

143| Parameter | Type | Description |144| Parameter | Type | Description |

144|-----------|------|-------------|145|-----------|------|-------------|

145| `instructions` | string | System prompt |146| `instructions` | string | System prompt |

146| `reasoning.effort` | `"high"` | `"none"` | optional | Controls whether the model uses reasoning. Defaults to `"high"`. Supported only with `grok-voice-latest` and `grok-voice-think-fast-1.0`. |147| `reasoning.effort` | `"high"` | `"none"` | optional | Controls whether the model uses reasoning. Defaults to `"high"`. |

147| `voice` | string | Voice selection: any built-in voice (e.g. `eve`, the default) or a [custom voice ID](/developers/model-capabilities/audio/custom-voices) (see [Available Voices](#available-voices)) |148| `voice` | string | Voice selection: any built-in voice (e.g. `eve`, the default) or a [custom voice ID](/developers/model-capabilities/audio/custom-voices) (see [Available Voices](#available-voices)) |

148| `tools` | array | Tools available to the voice agent. Supports `file_search`, `web_search`, `x_search`, `mcp`, and `function` types. See [Using Tools](#using-tools-with-grok-speech-to-speech-api). |149| `tools` | array | Tools available to the voice agent. Supports `file_search`, `web_search`, `x_search`, `mcp`, and `function` types. See [Using Tools](#using-tools-with-grok-speech-to-speech-api). |

149| `turn_detection.type` | string | null | `"server_vad"` for automatic detection, `null` for manual text turns |150| `turn_detection.type` | string | null | `"server_vad"` for automatic detection, `null` for manual text turns |

1619 1620 

1620### Step 3 — Model-Specific Best Practices1621### Step 3 — Model-Specific Best Practices

1621 1622 

1622#### `grok-voice-think-fast-1.0` (Recommended)1623#### `grok-voice-think-fast-2.0` (Recommended)

1623 1624 

1624This is the flagship voice model. Use `grok-voice-latest` for new integrations so your app tracks the current recommended model. When migrating:1625This is the newest voice model. Use `grok-voice-latest` for new integrations so your app tracks the current recommended model. When migrating:

1625 1626 

1626* **Simplify your system prompt.** The model is significantly more capable, so your prompt should be much shorter. Ask Grok to generalize your existing system prompt rather than porting it verbatim.1627* **Simplify your system prompt.** The model is significantly more capable, so your prompt should be much shorter. Ask Grok to generalize your existing system prompt rather than porting it verbatim.

1627* **Remove workaround prompting.** Prompt hacks and edge-case fixes needed for GPT models are unnecessary. Strip out instructions added solely to patch bugs or limitations of the previous model.1628* **Remove workaround prompting.** Prompt hacks and edge-case fixes needed for GPT models are unnecessary. Strip out instructions added solely to patch bugs or limitations of the previous model.

1629 1630 

1630> [!NOTE]1631> [!NOTE]

1631>1632>

1632> `grok-voice-latest` always points to the newest model (currently `grok-voice-think-fast-1.0`). Pin to a versioned model name in production for stability.1633> `grok-voice-latest` always points to the newest model (currently `grok-voice-think-fast-2.0`). Pin to a versioned model name in production for stability.

1633 1634 

1634## OpenAI Realtime API Compatibility1635## OpenAI Realtime API Compatibility

1635 1636 

4 4 

5The xAI Voice APIs offer a range of powerful voice capabilities, all powered by Grok, with enterprise-grade reliability and sub-second latency.5The xAI Voice APIs offer a range of powerful voice capabilities, all powered by Grok, with enterprise-grade reliability and sub-second latency.

6 6 

7## Speech to Speech API7## Speech to Speech

8 8 

9Build real-time, speech-to-speech voice agents over WebSockets, with low-latency turn-taking and tool use. For client-side apps, use [Ephemeral Tokens](/developers/model-capabilities/audio/ephemeral-tokens) to connect securely without exposing your API key.9Build real-time, speech-to-speech voice agents over WebSockets, with low-latency turn-taking and tool use. For client-side apps, use [Ephemeral Tokens](/developers/model-capabilities/audio/ephemeral-tokens) to connect securely without exposing your API key.

10 10 

39| grok-4.20-0309-non-reasoning | T0: 37, T1: 50, T2: 75, T3: 125, T4: 208 | T0: 10M, T1: 15M, T2: 25M, T3: 45M, T4: 85M |39| grok-4.20-0309-non-reasoning | T0: 37, T1: 50, T2: 75, T3: 125, T4: 208 | T0: 10M, T1: 15M, T2: 25M, T3: 45M, T4: 85M |

40| grok-build-0.1 | T0: 37, T1: 50, T2: 75, T3: 125, T4: 208 | T0: 10M, T1: 15M, T2: 25M, T3: 45M, T4: 85M |40| grok-build-0.1 | T0: 37, T1: 50, T2: 75, T3: 125, T4: 208 | T0: 10M, T1: 15M, T2: 25M, T3: 45M, T4: 85M |

41| grok-4.20-multi-agent-0309 | T0: 9, T1: 12, T2: 18, T3: 31, T4: 56 | T0: 2.5M, T1: 3.7M, T2: 6.2M, T3: 11M, T4: 21M |41| grok-4.20-multi-agent-0309 | T0: 9, T1: 12, T2: 18, T3: 31, T4: 56 | T0: 2.5M, T1: 3.7M, T2: 6.2M, T3: 11M, T4: 21M |

42| grok-imagine-image | 5 | — |

43| grok-imagine-image-quality | 5 | — |42| grok-imagine-image-quality | 5 | — |

43| grok-imagine-image | 5 | — |

44| grok-imagine-video-1.5 | 10 | — |44| grok-imagine-video-1.5 | 10 | — |

45| grok-imagine-video | 10 | — |45| grok-imagine-video | 10 | — |

46 46 

14 14 

15* `session` (object | null) — Optional initial session configuration to bind to the client secret. This JSON value is stored alongside the secret and applied when the WebSocket connection opens.15* `session` (object | null) — Optional initial session configuration to bind to the client secret. This JSON value is stored alongside the secret and applied when the WebSocket connection opens.

16 16 

17 * `model` ("grok-voice-latest" | "grok-voice-think-fast-1.0") — Model to use for the session. Use grok-voice-latest for the best experience.17 * `model` ("grok-voice-latest" | "grok-voice-think-fast-2.0" | "grok-voice-think-fast-1.0") — Model to use for the session. Use grok-voice-latest for the best experience.

18 18 

19 * `reasoning` (object) — Reasoning settings for models that support them.19 * `reasoning` (object) — Reasoning settings for models that support them.

20 20 

21 * `effort` ("high" | "none") — Controls whether the model uses reasoning. Defaults to \`high\`. Supported only with \`grok-voice-latest\` and \`grok-voice-think-fast-1.0\`.21 * `effort` ("high" | "none") — Controls whether the model uses reasoning. Defaults to \`high\`.

22 22 

23### Response Body23### Response Body

24 24 

207 207 

208* `model` (string, optional, default: grok-voice-latest) — Model to use for the session. Ignored when \`call\_id\` is provided because the session is bound to the inbound SIP call. Use grok-voice-latest for the best experience on direct WebSocket sessions.208* `model` (string, optional, default: grok-voice-latest) — Model to use for the session. Ignored when \`call\_id\` is provided because the session is bound to the inbound SIP call. Use grok-voice-latest for the best experience on direct WebSocket sessions.

209 209 

210* `reasoning.effort` (string, optional, default: high) — Controls whether the model uses reasoning. Defaults to \`high\`. Supported only with grok-voice-latest and grok-voice-think-fast-1.0.210* `reasoning.effort` (string, optional, default: high) — Controls whether the model uses reasoning. Defaults to \`high\`.

211 211 

212### Client Messages212### Client Messages

213 213