AI Models
Providers, model types, and how Assistant Core picks a model.
Assistant Core supports multiple AI models for chat, voice, and realtime conversations. Every plan can use every active model; usage is billed in credits, so more capable models simply cost more credits.
Where Models Are Selected
| Surface | What it controls |
|---|---|
| Web App → chat header | The Chat model for text messages, attachments, tool use, and Knowledge Base retrieval |
| Web App → Settings → Voice | Personal Voice Agent model and voice preferences for Standard mode and Realtime mode |
| Admin Panel → Assistant Settings → Models | The assistant's own Chat, voice, and realtime models, voices, and spoken language |
| Admin Panel → Assistant Settings | Assistant-level voice enablement, recording policy, and Stream Response Text |
| Platform/Admin catalog | Model catalog, voice lists, the platform default, and display order per model type |
The Chat model and Voice Agent models are separate choices. Changing the Chat model in the header does not change VAD, ASR, TTS, TTS voice, Realtime model, or Realtime voice in Settings → Voice.
Providers
| Provider | Supports |
|---|---|
| OpenAI | Chat, speech to text, realtime voice |
| Anthropic | Chat |
| Google AI | Chat, realtime voice |
| X.AI | Chat, text to speech, speech to text, realtime voice |
| DeepSeek | Chat |
| Xiaomi | Chat |
| ElevenLabs | Text to speech, speech to text |
| Soniox | Text to speech, speech to text |
| Local/Open Source | Voice activity detection |
Model Types
| Type | What it is used for |
|---|---|
| Chat | Answers questions, reasons through tasks, uses tools, and handles multimodal content when supported |
| Voice Activity Detection | Detects when a user starts or stops speaking |
| Speech to Text | Transcribes spoken audio into text |
| Text to Speech | Turns text into spoken audio |
| Realtime Voice | Low-latency two-way voice conversations |
Voice Models
Voice Activity Detection
| Provider | Models |
|---|---|
| Local/Open Source | Silero VAD |
Speech To Text
| Provider | Models |
|---|---|
| OpenAI | GPT-4o Transcribe, GPT-4o Mini Transcribe |
| ElevenLabs | ElevenLabs Scribe v2 |
| Soniox | Soniox STT Realtime v5 |
| X.AI | Grok STT |
Text To Speech
| Provider | Models |
|---|---|
| ElevenLabs | ElevenLabs Flash v2.5 |
| X.AI | Grok 2 TTS |
| Soniox | Soniox TTS Realtime v1 |
Realtime Voice
| Provider | Models |
|---|---|
| OpenAI | GPT Realtime 2.1, GPT Realtime 2.1 Mini, GPT Realtime 2, GPT Realtime 1.5 |
| Google AI | Gemini 3.8 Live |
| X.AI | Grok Voice Agent |
Voices by Model
Each TTS and Realtime model has its own set of voices. When you switch models, the list of available voices changes accordingly.
Text to Speech (Standard Mode)
| Model | Voices | Default |
|---|---|---|
| Grok 2 TTS (X.AI) | Ara, Eve, Leo, Rex, Sal | Ara |
| ElevenLabs Flash v2.5 | Aria, Brian, Jessica, Chris, Lily, Alice, George, Huyền, Jade, Nhật Nam, Adam | — (first in list) |
| Soniox TTS Realtime v1 | Adrian, Maya, Daniel, Noah, Nina, Emma, Jack, Claire, Grace, Owen, Mina, Kenji | Adrian |
Realtime Voice
| Model | Voices | Default |
|---|---|---|
| Grok Voice Agent (X.AI) | Ara, Eve, Leo, Rex, Sal | Ara |
| GPT Realtime 2.1 / 2.1 Mini / 2 / 1.5 (OpenAI) | Alloy, Ash, Ballad, Coral, Echo, Sage, Shimmer, Verse, Marin, Cedar | Verse |
| Gemini 3.8 Live (Google AI) | Kore, Puck, Charon, Fenrir, Aoede | — (first in list) |
Vietnamese (vi-VN) voices are currently available only on ElevenLabs: Huyền, Jade (female) and Nhật Nam (male). The other voices are English (en-US/en-GB) but can still read Vietnamese text.
Chat Models
| Provider | Models |
|---|---|
| OpenAI | GPT-6 Luna, GPT-6.1 Sol |
| Anthropic | Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5 |
| Google AI | Gemini 3.8 Flash, Gemini 3.5 Flash Lite |
| X.AI | Grok 4.7, Grok 4.3 |
| DeepSeek | DeepSeek Flash |
| Xiaomi | MiMo V2.6 Flash, MiMo V2.6 Pro |
OpenAI token rates
Assistant Core converts provider cost to credits using your plan's billing multiplier. These are the underlying OpenAI Standard rates in USD per 1 million tokens:
| Model | Input context | Input | Cached input | Cache write | Output |
|---|---|---|---|---|---|
| GPT-6 Luna | ≤272K | $0.10 | $0.01 | $0.125 | $0.50 |
| GPT-6 Luna | >272K | $0.20 | $0.02 | $0.25 | $0.75 |
| GPT-6.1 Sol | ≤272K | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6.1 Sol | >272K | $4.00 | $0.20 | $5.00 | $15.00 |
For GPT-6.1 Sol and GPT-6 Luna, more than 272K input tokens switches the full request to the long-context rates. Each model supports up to 922K input tokens plus 128K output tokens within its 1.05M context window.
How a Model Is Chosen
Plans do not restrict models. For each model type, the first available choice wins:
- The model picked for the request — the Web App chat header, or
modelin the API. - Your own preference in Settings → Voice — applies when you pay for the usage (Web App and your devices).
- The assistant's configuration, set by its owner in the Admin Panel.
- The platform default below.
An empty choice is labeled with what it resolves to, for example Assistant default (Grok 2 TTS). Widget and API usage is billed to the assistant owner, so it always uses the assistant's configuration. Before each turn, your credit balance must cover a typical turn on the chosen model.
Default Models
When neither you nor the assistant owner picks a model, Assistant Core uses these platform defaults:
| Type | Default Model |
|---|---|
| Chat (LLM) | DeepSeek Flash |
| Text to Speech | Grok 2 TTS |
| Speech to Text | Soniox STT Realtime v5 |
| Voice Activity Detection | Silero VAD |
| Realtime Voice | Gemini 3.8 Live |
Defaults and available models are the same on every plan. Plans differ in monthly credits and features, not in model access.
Each AI operation consumes credits from your wallet. More expensive models cost more credits. See Credits for details.