AI Models & Selection
SOHAM routes your queries across 13+ specialized models from Groq, Google, HuggingFace, and OpenRouter. Here's what's available and when to use each.
Auto Mode — Recommended for Most Users
SOHAM analyzes your query and routes it to the most capable model automatically
Math problems
Routed to math-capable models (Qwen, Llama)
Code questions
Routed to DeepSeek V3.2 or Llama 3.3 70B
Image tasks
Routed to Gemini multimodal models
Web queries
Auto-triggers DuckDuckGo search + AI synthesis
General chat
Routed to fast conversational models
Document tasks
Routed to large-context models
Available Models
General Purpose
Fast, versatile models for everyday tasks
Llama 3.1 8B Instant
Groq — Ultra-fast inference, great for quick answers and general chat
Llama 3.2 1B InstructFree
HuggingFace — Compact free model, good for lightweight tasks
Llama 3.1 70B InstructFree
HuggingFace — Large free model with strong reasoning
Qwen 2.5 7B InstructFree
HuggingFace — Strong multilingual model, excellent at math and code
Coding Specialists
Models fine-tuned for programming, debugging, and architecture
DeepSeek V3.2
HuggingFace — State-of-the-art coding model — best for complex programming tasks
Conversational
Models optimized for natural dialogue
RNJ-1 Instruct
HuggingFace — Efficient conversational model with good personality
Gemini 2.5 Flash Lite
Google — Lightweight Google model, fast and helpful
Multimodal (Vision)
Models that understand and analyze images
Gemini 2.5 Flash
Google — Google's latest multimodal model — best for image analysis
Gemini Flash Latest
Google — Latest Gemini Flash with improved vision capabilities
Free / OpenRouter
Completely free models with no usage limits
OpenRouter Auto (Free)Free
OpenRouter — Automatically selects the best available free model from OpenRouter
How to Select a Model
Desktop
- 1. Click the Settings icon (⚙️) in the header
- 2. Find the "AI Model" dropdown
- 3. Select "Auto" or a specific model
- 4. Settings save automatically
Mobile
- 1. Tap the model button in the chat header
- 2. A bottom sheet opens with all models
- 3. Tap to select your preferred model
- 4. Sheet closes and model is applied
💡 Model Selection Tips
For coding tasks
Use DeepSeek V3.2 for complex programming, debugging, and architecture questions.
For math problems
Auto mode routes to Qwen 2.5 or Llama models which handle math well.
For image analysis
Use Gemini 2.5 Flash for image understanding and visual Q&A.
For speed
Groq's Llama 3.1 8B Instant is the fastest model available.
For free unlimited use
HuggingFace models (Llama, Qwen) and OpenRouter Auto have no usage limits.
For general chat
Auto mode or RNJ-1 Instruct work best for everyday conversations.
Model Providers
Groq
Ultra-fast inference with streaming
HuggingFace
Open-source models, many free
Advanced multimodal capabilities
OpenRouter
Aggregated access to many models