Skip to content

Models

This page provides information about the AI models supported by VT.ai and how to use them effectively.

Supported Models

VT.ai integrates with multiple AI providers and supports a wide range of models:

OpenAI Models

  • GPT-o1 (o1): High-performance general purpose model
  • GPT-o1 Mini (o1-mini): Compact version of GPT-o1
  • GPT-o1 Pro (o1-pro): Enhanced version with advanced capabilities
  • GPT-o3 Mini (o3-mini): Compact, efficient model for everyday tasks
  • GPT-4.5 Preview (gpt-4.5-preview): Preview of next-generation capabilities
  • GPT-4o (4o): Advanced vision and multimodal capabilities

Anthropic Models

  • Claude 3.5 Sonnet (c3.5-sonnet): Balanced performance and efficiency
  • Claude 3.7 Sonnet (sonnet): Advanced reasoning capabilities
  • Claude 3.5 Haiku (c3.5-haiku): Fast, efficient model for common tasks
  • Claude 3 Opus (opus): Highest capability model for complex tasks

Google Models

  • Gemini 1.5 Pro (gemini-1.5-pro): Advanced multimodal capabilities
  • Gemini 1.5 Flash (gemini-1.5-flash): Fast, efficient model
  • Gemini 2.0 (gemini-2.0): Advanced reasoning model
  • Gemini 2.5 Pro (gemini-2.5-pro): Latest Google model with enhanced capabilities
  • Gemini 2.5 Flash (gemini-2.5-flash): Fast version of Gemini 2.5

DeepSeek Models

  • DeepSeek-Coder (deepseek-coder): Specialized for coding tasks
  • DeepSeek Chat (deepseek-chat): General conversation model
  • DeepSeek R1 Series (deepseek-r1): Next-generation reasoning models (multiple sizes)

Groq Models

  • Llama 4 Scout 17b Instruct: Fast inference of Llama 4 Scout via Groq
  • Llama 3 8b/70b: Optimized versions of Llama 3 on Groq's infrastructure
  • Mixtral 8x7b: Fast inference of Mixtral model

Cohere Models

  • Command: General purpose instruction model
  • Command-R: Enhanced reasoning capabilities
  • Command-Light: Lightweight, efficient model
  • Command-R-Plus: Advanced reasoning with extended capabilities

OpenRouter Integration

VT.ai supports many models through OpenRouter, including:

  • Qwen Models: Qwen 2.5 VL 32B, Qwen 2.5 Coder 32B, etc.
  • Mistral Models: Mistral Small 3.1 24B and others
  • Additional proprietary and open models

Local Models (via Ollama)

Run models locally for privacy and offline use:

  • Llama 3: Multiple sizes (8B, 70B)
  • DeepSeek R1: Various sizes (1.5B, 7B, 8B, 14B, 32B, 70B)
  • Qwen2.5-coder: Multiple versions (7b, 14b, 32b)
  • Mistral: Various versions
  • Many other open source models

Model Selection

You can select models in several ways:

  1. Command Line:
vtai --model sonnet
  1. UI Settings:
  2. Use the model selector in the settings menu

  3. Change models during a conversation

  4. Dynamic Routing:

  5. Allow VT.ai to automatically select the best model for your query
  6. Enable in settings with "Use Dynamic Conversation Routing"

Model Capabilities

Different models have different capabilities:

Vision-Capable Models

For analyzing images and visual content:

  • GPT-4o
  • Gemini 1.5 Pro/Flash
  • Gemini 2.5 Pro/Flash
  • Claude 3 Sonnet/Opus
  • Llama 3.2 Vision
  • Qwen 2.5 VL

TTS-Capable Models

For text-to-speech generation:

  • GPT-4o mini TTS
  • TTS-1
  • TTS-1-HD
  • Various voice options: alloy, echo, fable, onyx, nova, shimmer

Image Generation Models

For creating images from text descriptions:

  • DALL-E 3: OpenAI's image generation model
  • GPT-Image-1: Advanced image generation model with customizable settings
  • Supports transparent backgrounds
  • Multiple output formats (PNG, JPEG, WEBP)
  • Customizable dimensions (square, landscape, portrait)
  • Quality settings (standard, high)
  • Advanced compression options (0-100 for webp/jpeg)
  • HD options for higher quality outputs

Reasoning-Enhanced Models

VT.ai supports special "thinking mode" with these models:

  • DeepSeek Reasoner
  • DeepSeek R1 series
  • Qwen 2.5 models
  • Claude 3 models
  • GPT-4o

Performance Considerations

When choosing models, consider these factors:

  • Speed: Models like GPT-o3 Mini, Groq-accelerated models, and Claude 3 Haiku are faster
  • Quality: Models like GPT-o1, GPT-4o, and Claude 3 Opus offer higher quality
  • Cost: Smaller models generally cost less to use
  • Multimodal Needs: Only some models support image analysis
  • Local Computation: Ollama models run locally but require more resources
  • API Availability: Some models may require specific API keys

API Key Configuration

For most models, you'll need to configure the appropriate API keys:

# OpenAI models (o1, o3-mini, 4o, etc.)
export OPENAI_API_KEY="sk-your-key-here"

# Anthropic models (sonnet, opus, etc.)
export ANTHROPIC_API_KEY="sk-ant-your-key-here"

# Google models (gemini series)
export GEMINI_API_KEY="your-key-here"

# DeepSeek models
export DEEPSEEK_API_KEY="your-key-here"

# Groq models
export GROQ_API_KEY="your-key-here"

# Cohere models
export COHERE_API_KEY="your-key-here"

# OpenRouter (for access to multiple providers)
export OPENROUTER_API_KEY="your-key-here"

You can also set these keys when starting VT.ai:

vtai --api-key openai=sk-your-key-here

For more details on configuration, see the Configuration page.