Model Selection Guide

Choose the right AI model for your needs.

Available Models & Providers

Astronox supports 6 AI providers with a combined total of 12 distinct models.

Gemini Models (Google)

Gemini 2.5 Flash (Recommended)

Best for: Most everyday tasks

Strengths:

• ⚡ Very fast (1–2 seconds)
• 💰 Free tier (60 req/min, 1500 req/day)
• 🎯 Excellent balance of speed & quality
• 📸 Good image analysis
• 💻 Strong code generation

Specs:

• Context: 128k tokens (~100k words)
• Speed: ⚡⚡⚡
• Quality: ⭐⭐⭐⭐
• Cost: 💰 Free / $0.075 per 1M input tokens (paid)

Use when:

• General questions and productivity
• File management
• Image analysis
• Code generation
• Quick responses needed

Gemini 2.5 Flash Lite

Best for: Simple tasks, maximum speed

Strengths:

• ⚡⚡ Fastest (<1 second)
• 💰💰 Extremely low cost
• 🔋 Efficient for simple queries

Specs:

• Context: 128k tokens
• Speed: ⚡⚡⚡⚡
• Quality: ⭐⭐⭐
• Cost: 💰 Free / $0.0375 per 1M input tokens

Best for:

• Brief summaries
• Simple classifications
• Quick fact lookups
• Saving on quota

Gemini 2.5 Pro

Best for: Complex reasoning, research

Strengths:

• 💪 Most powerful Gemini
• 🧠 Better reasoning than Flash
• 📊 Handles large documents well
• ✏️ Superior writing quality

Specs:

• Context: 128k tokens
• Speed: ⚡⚡
• Quality: ⭐⭐⭐⭐⭐
• Cost: 💰💰 $1.25 per 1M input tokens (requires billing)

Use when:

• Complex analysis
• Research synthesis
• Writing tasks
• Advanced reasoning needed

OpenAI Models (GPT-5.2 Series)

GPT-5.2 (Recommended)

Best for: All-purpose tasks, professional use

Strengths:

• 🎯 Versatile for any task
• 💪 Powerful reasoning
• ⚡ Fast responses
• 💻 Excellent code generation
• 🧠 Advanced understanding

Specs:

• Context: 128k tokens
• Speed: ⚡⚡
• Quality: ⭐⭐⭐⭐⭐
• Cost: 💰💰 $3 per 1M input tokens

Use when:

• Production tasks
• Complex code
• Professional writing
• Advanced reasoning
• When you need the best

GPT-5.2-Codex

Best for: Code generation and optimization

Strengths:

• 💻 Optimized for code
• 🚀 Better at code patterns
• 🔍 Superior code review
• ✨ Generates efficient code

Specs:

• Context: 128k tokens
• Speed: ⚡⚡
• Quality: ⭐⭐⭐⭐⭐ (for code)
• Cost: 💰💰 $3 per 1M input tokens

Use when:

• Generating code
• Debugging
• Code optimization
• Architecture design
• Technical documentation

Claude Models (Anthropic)

Claude Haiku 4.5 (Recommended)

Best for: Fast, efficient reasoning

Strengths:

• ⚡ Extremely fast
• 💡 Surprisingly capable
• 💰 Most affordable
• 🎯 Good reasoning for the price
• 🧠 Better understanding than Flash Lite

Specs:

• Context: 200k tokens (~150k words)
• Speed: ⚡⚡⚡
• Quality: ⭐⭐⭐⭐
• Cost: 💰 $0.80 per 1M input tokens

Use when:

• Fast reasoning needed
• Budget-conscious
• General assistance
• Most daily tasks

Claude Sonnet 4.5

Best for: Advanced reasoning and nuance

Strengths:

• 🧠 Superior reasoning
• ✍️ Better writing quality
• 🎯 Excellent understanding of nuance
• 📚 Great with long documents
• 🔍 Detailed analysis

Specs:

• Context: 200k tokens (~150k words)
• Speed: ⚡⚡
• Quality: ⭐⭐⭐⭐⭐
• Cost: 💰💰 $3 per 1M input tokens

Use when:

• Complex reasoning needed
• Writing and content
• Nuanced analysis
• Research synthesis
• When quality matters most

Azure OpenAI Models

Same as OpenAI (GPT-5.2 series), deployed on Azure infrastructure.

Benefits:

• 🔒 Enterprise security
• 📍 Regional deployment
• 🏢 HIPAA/compliance ready
• ⚙️ Custom fine-tuning available
• 🔐 Your data stays in your region

Auto Provider Mode

Select "Auto" to let Astronox choose:

1. Simple queries → Gemini Flash (fastest, free)
2. Moderate tasks → Gemini Pro (balanced)
3. Complex/reasoning → Claude Sonnet 4.5 (best reasoning)
4. Code generation → GPT-5.2-Codex (code optimized)
5. Fallback → Silent automatic retry with different provider

Benefits:

• ✨ Best model for each task
• 💰 Optimized cost
• ⚡ Optimized speed
• 🎯 Optimized quality

Quick Comparison

Cost Comparison (Per 1M Tokens)

How to Switch Models

1. Open Astronox Settings
2. Select Provider Tab (Gemini, OpenAI, Claude, etc.)
3. Choose Model from dropdown (if available)
4. Click Save
5. Next message uses new model

Tips for Better Results

Choose Model for Task

• Simple task? → Gemini Flash or Haiku (fast, cheap)
• Complex task? → Gemini Pro, Sonnet, or GPT-5.2 (better reasoning)
• Code task? → GPT-5.2-Codex (code optimized)

Use Auto Mode for Best Results

Auto Mode intelligently picks the best model based on what you ask.

• Switch to Flash or Flash Lite
• Reduce conversation history (new chat)
• Simplify your request
• Check internet speed

Next: Configure your Settings to customize your Astronox experience.

• ✅ Good for straightforward tasks

Specs:

• Max context: 32k tokens
• Speed: ⚡⚡⚡ Excellent++
• Quality: ⭐⭐⭐ Good
• Cost: 💰💰 Even cheaper than Flash

Use when:

• You need instant answers
• Simple file operations
• Quick questions
• Basic searches
• When cost is a major concern

Not ideal for:

• Complex multi-step tasks
• Detailed code generation
• Deep analysis
• Large context requirements

Gemini 2.5 Pro

Best for: Complex tasks, maximum quality

Strengths:

• 🎯 Highest reasoning quality
• 📚 Massive context (2 million tokens!)
• 🧠 Best for complex problems
• 📊 Superior analysis
• 💻 Advanced code generation
• 🖼️ Best image understanding

Specs:

• Max context: 2M tokens (~1.5M words)
• Speed: ⚡⚡ Good (3-5 seconds)
• Quality: ⭐⭐⭐⭐⭐ Excellent
• Cost: 💰💰 Higher (still generous free tier)

Use when:

• Complex multi-step workflows
• Large codebases analysis
• Detailed documentation review
• Advanced problem-solving
• Working with massive context
• Highest quality output needed

Example tasks:

• "Analyze my entire project and suggest architecture improvements"
• "Create a comprehensive backup and organization system"
• "Review these 50 files and find optimization opportunities"
• "Build a complete deployment pipeline"

Trade-offs:

• Slower responses (worth it for complex tasks)
• Higher API costs (but still reasonable)
• May be overkill for simple tasks

MintAI Models

Devstral 2 (Pro)

Best for: Pro subscribers, code-specialized tasks

Strengths:

• 💻 Optimized for code (Mistral AI's coding model)
• 🛠️ Excellent tool integration
• ⚡ Fast responses
• 🎯 Strong technical accuracy
• 🔐 Automatic key management (Pro)

Specs:

• Max context: 128k tokens
• Speed: ⚡⚡⚡ Excellent
• Quality: ⭐⭐⭐⭐ Very Good (especially code)
• Cost: 💰 Included with Pro subscription

Use when:

• You're a Pro subscriber
• Heavy code generation/review
• Technical automation tasks
• Development workflows
• Want alternative to Gemini

Availability:

• Requires active Pro subscription
• API key auto-fetched by Astronox
• No manual configuration needed

Model Comparison Table

Choosing the Right Model

Decision Tree

Need maximum quality or huge context?
├─ Yes → Gemini Pro
└─ No ↓

Working primarily with code?
├─ Yes (+ Pro) → Devstral 2
├─ Yes (no subscription) → Gemini Flash
└─ No ↓

Need instant responses? Simple tasks only?
├─ Yes → Gemini Flash Lite
└─ No → Gemini Flash (default)

By Task Type

📁 File Management

• Best: Gemini Flash
• Alternative: Flash Lite (simple), Pro (complex workflows)

💻 Code Generation

• Best: Devstral 2 (Pro) or Gemini Pro
• Alternative: Gemini Flash (very capable)

🖼️ Image Analysis

• Best: Gemini Pro
• Alternative: Gemini Flash (usually sufficient)

🔍 Search & Analysis

• Best: Gemini Flash (most cases), Pro (large datasets)

🤖 Automation Scripts

• Best: Devstral 2 or Gemini Pro
• Alternative: Gemini Flash

📊 Data Processing

• Best: Gemini Pro (large data), Flash (typical)

💬 General Questions

• Best: Gemini Flash Lite or Flash

🧠 Complex Problem-Solving

• Best: Gemini Pro
• Alternative: Devstral 2 (technical), Flash (simpler)

Switching Models

In Settings

1. Click Settings (⚙️) → General tab
2. Find Model Selection dropdown
3. Select your preferred model:
   • Gemini 2.5 Flash (default)
   • Gemini 2.5 Flash Lite
   • Gemini 2.5 Pro
   • Devstral 2 (if Pro active)
4. Click Save Settings

Via Chat

"Switch to Gemini Pro"
"Use the fastest model"
"Change to Devstral 2"

Per-Conversation

Model selection applies to new messages in current conversation. You can:

• Switch mid-conversation (new responses use new model)
• Start new chat to reset context with different model

Performance Characteristics

Response Time

Typical latency (simple request):

• Flash Lite: 0.5-1s
• Flash: 1-2s
• Devstral 2: 1-2.5s
• Pro: 2-5s

Complex multi-step task:

• Flash Lite: Not recommended
• Flash: 5-15s
• Devstral 2: 5-20s
• Pro: 10-30s

Token Limits

What are tokens?
Tokens ≈ words/4. Example: 1000 words ≈ 1333 tokens.

Model limits:

• Flash Lite: 32k (~24k words) - about 40 pages
• Flash: 128k (~96k words) - about 150 pages
• Devstral 2: 128k (~96k words) - about 150 pages
• Pro: 2M (~1.5M words) - entire novel + codebases

What counts toward limit:

• Your conversation history in current chat
• System prompt (~2k tokens)
• Your message + attachments
• AI's responses

When you hit limit:

• Older messages get truncated (auto-managed)
• Or start new conversation

Cost Considerations

Free Tier Limits (Gemini)

Flash & Flash Lite:

• 60 requests per minute
• 1,500 requests per day
• More than enough for typical use

Pro:

• 10 requests per minute
• 500 requests per day
• Still generous for power users

If you exceed free tier:

• Set up billing in Google AI Studio
• Pay per token (very affordable)
• Flash: ~$0.001 per request (typical)
• Pro: ~$0.005 per request (typical)

Pro (MintAI)

Subscription model:

• Fixed monthly/annual fee
• Unlimited requests (within reason)
• Access to multiple premium models
• No per-token billing

Cost comparison:

• Heavy users (100+ requests/day): Subscription may be cheaper
• Light users (<50 requests/day): Free tier likely sufficient

Quality Differences Explained

Understanding "Quality"

What varies between models:

1. Reasoning depth: How well they understand complex requests
2. Accuracy: Correctness of information and code
3. Context awareness: Remembering earlier conversation
4. Instruction following: Doing exactly what you asked
5. Creativity: Generating novel solutions

Real-World Examples

Simple task: "List files in Downloads"

• All models: Nearly identical results
• Flash Lite is faster, no quality loss

Medium task: "Write a Python script to organize files by date"

• Flash/Devstral: Great scripts, maybe minor issues
• Pro: More robust error handling, better comments

Complex task: "Analyze my entire codebase and refactor for performance"

• Flash: Good insights, may miss patterns
• Pro: Comprehensive analysis, deeper understanding
• Devstral: Strong on code specifics

Image task: "Describe everything in this complex diagram"

• Flash: Identifies main elements
• Pro: Detailed analysis, relationships, subtle details

Accuracy & Reliability

Code Generation Accuracy

Rankings (by testing):

1. Gemini Pro (98% working code)
2. Devstral 2 (97% working code)
3. Gemini Flash (95% working code)
4. Flash Lite (90% working code)

For production code: Use Pro or Devstral 2
For prototypes/scripts: Flash is fine
For learning examples: Any model works

Factual Accuracy

All models can occasionally:

• Hallucinate (make up information)
• Misunderstand context
• Provide outdated info

Mitigation:

• Verify critical information
• Ask model to cite sources
• Cross-check important decisions
• Use Pro for high-stakes tasks

Advanced Model Features

Function/Tool Calling

All models support Astronox's tool system:

• File operations
• System control
• Multi-step planning

Best at tool calling:

1. Gemini Pro (most accurate)
2. Gemini Flash (very reliable)
3. Devstral 2 (good, code-focused)
4. Flash Lite (basic support)

Multi-Modal (Vision)

Image understanding:

• Pro: Best detail and accuracy
• Flash: Very good for most images
• Devstral 2: Good, less specialized
• Flash Lite: Basic recognition

Supported image types: JPG, PNG, GIF, WebP

Code-Specific Features

Devstral 2 advantages:

• Optimized for code completion
• Better understanding of frameworks
• Strong at refactoring patterns
• Excellent at debugging

Best Practices

✅ Start with Flash

Default model (Flash) handles 90% of tasks excellently.

✅ Upgrade to Pro for:

• Large file analysis
• Complex multi-step workflows
• Detailed code architecture
• When Flash gives unsatisfactory results

✅ Use Flash Lite for:

• Quick questions
• Simple file listings
• Basic searches
• When speed matters most

✅ Try Devstral 2 if:

• You're a Pro subscriber
• Heavy code-focused workflows
• Want alternative perspective
• Exploring different model capabilities

✅ Monitor Your Usage

• Settings → API Keys → View Usage (future)
• Track if approaching free tier limits
• Upgrade to paid tier if needed

Troubleshooting Model Issues

"Rate limit exceeded"

Solution:

• You've hit free tier limits
• Wait for rate limit reset (1 minute or 1 day)
• Or enable billing in Google AI Studio
• Or switch to different model/provider

"Model not available"

Solutions:

• Devstral 2: Check Pro subscription is active
• Gemini: Verify API key is valid
• Check internet connection

"Poor quality responses"

Try:

1. Switch to higher-quality model (Pro)
2. Provide more context in your prompt
3. Break complex tasks into simpler steps
4. Start new conversation (reset context)

"Too slow"

Solutions:

• Switch to Flash or Flash Lite
• Reduce conversation history (new chat)
• Simplify your request
• Check internet speed

Next: Configure your Settings to customize your Astronox experience.