Understanding AI Models: How to Choose the Right One in Archevi

Archevi gives you access to **6 different AI models** for answering questions about your documents. But which one should you use? Here's a practical guide.

The 6 Models Available

### Groq Models (Lightning-Fast)

**Groq Llama 3.3 70B**

Best for: General questions, fastest responses
Speed: ~500 tokens/second (near-instant)
Cost: Lowest
Quality: Good for most queries

**Groq Llama 4 Scout (9B)**

Best for: Simple lookups, quick facts
Speed: Extremely fast
Cost: Very low
Quality: Great for straightforward questions

**Groq Llama 4 Maverick (12B)**

Best for: Balanced speed and quality
Speed: Very fast
Cost: Low
Quality: Better reasoning than Scout

### Cohere Models (Highest Quality)

**Cohere Command A**

Best for: Complex reasoning, detailed analysis
Speed: Moderate
Cost: Higher
Quality: Excellent for nuanced questions

**Cohere Command R**

Best for: Research-heavy queries with many sources
Speed: Moderate
Cost: Moderate
Quality: Optimized for RAG (Retrieval-Augmented Generation)

**Cohere Command R+**

Best for: Maximum accuracy and depth
Speed: Slower
Cost: Highest
Quality: Best available

When to Use Each Model

### Use Groq Llama 3.3 70B (Default) When:

You want fast, reliable answers
Your question is straightforward ("What's our deductible?")
You're doing multiple queries and want to conserve AI budget
You're not sure which model to pick

### Use Groq Llama 4 Scout When:

You need the absolute fastest response
Your question is very simple ("When does this expire?")
You're on a tight AI budget

### Use Groq Llama 4 Maverick When:

You want better reasoning than Scout
Your question requires some analysis
You still want very fast responses

### Use Cohere Command R When:

Your question requires searching multiple documents
You need the AI to synthesize information from several sources
Accuracy is more important than speed

### Use Cohere Command A When:

Your question is complex or nuanced
You need detailed analysis
The stakes are high (legal, medical, financial decisions)

### Use Cohere Command R+ When:

You need the absolute best answer
Your question is very complex
Cost is not a concern
This is a critical decision

Real-World Examples

**Question:** "What's our home insurance deductible?" **Best Model:** Groq Llama 3.3 70B (default) - simple lookup, fast answer

**Question:** "Compare our car insurance options and tell me which one covers rental cars." **Best Model:** Cohere Command R - needs to search multiple documents and synthesize

**Question:** "What medical treatments did Dr. Smith recommend over the past 2 years, and are there any conflicting recommendations?" **Best Model:** Cohere Command A or R+ - requires deep analysis across time

**Question:** "When does my passport expire?" **Best Model:** Groq Llama 4 Scout - extremely simple, instant answer

How Model Selection Affects Your AI Budget

Each plan includes an AI budget:

**Trial:** $1/month
**Starter:** $3/month
**Family:** $8/month
**Family Office:** $25/month

Approximate costs per query:

Groq models: $0.001-0.002 per query
Cohere Command R: $0.005-0.01 per query
Cohere Command R+: $0.01-0.02 per query

**Strategy for budget management:** 1. Use Groq models (default) for 90% of queries 2. Switch to Cohere R for complex research queries 3. Reserve Cohere R+ for critical decisions only

How to Change Models

In the Archevi chat interface: 1. Click the **Model Selector** dropdown (top of chat) 2. Choose your preferred model 3. Your selection is saved for future queries

The default is **Groq Llama 3.3 70B** - optimized for speed and cost.

Coming Soon: Intelligent Model Routing

In v0.7.0, we're adding **automatic model selection** - Archevi will analyze your question and choose the best model for you. Simple questions → fast models. Complex analysis → powerful models.

Questions?

Still not sure which model to use? **Start with the default (Groq Llama 3.3 70B).** It's fast, accurate, and handles 90% of queries perfectly.

If you're not satisfied with the answer, try Cohere Command R for deeper analysis.