Model Selection

Choosing the right AI model tier for your task

Selecting the appropriate model tier balances quality, speed, and cost. This guide helps you make the right choice for each use case.

Model Tiers Overview

Tier	Primary Use	Speed	Cost	Capability
Low	Simple classification	Fastest	Lowest	Basic reasoning
Medium	Standard analysis	Balanced	Moderate	Good reasoning
High	Complex reasoning	Slower	Higher	Advanced reasoning

Tier Characteristics

Low Tier

Best for:

Binary yes/no decisions
Simple classification
Keyword extraction
Basic sentiment (positive/negative)
High-volume, simple tasks

Capabilities:

Fast response times
Good for straightforward tasks
Limited multi-step reasoning
May struggle with nuance

Example tasks:

text

- Is this meeting urgent? (boolean)
- Classify: sales/support/other (string)
- Extract mentioned company names (string_list)

Medium Tier (Default)

Best for:

Meeting summarization
Action item extraction
Content generation
Standard analysis
Most common workflows

Capabilities:

Balanced speed and quality
Good reasoning ability
Handles most business tasks
Reliable structured outputs

Example tasks:

text

- Summarize this meeting (string)
- Extract action items (string_list)
- Generate follow-up email (string)
- Identify key discussion points (string_list)

High Tier

Best for:

Complex analysis
Multi-factor decisions
Nuanced interpretation
Research tasks with tools
Strategic recommendations

Capabilities:

Advanced reasoning
Handles ambiguity well
Multi-step problem solving
Best output quality

Example tasks:

text

- Analyze deal risk factors (string)
- Research company with web search (string)
- Develop account strategy (string)
- Identify subtle sentiment shifts (string)

Selection Matrix

Use this matrix to choose the right tier:

Task Type	Complexity	Recommended Tier
Yes/No decision	Low	Low
Simple classification	Low	Low
Count items	Low	Low
Basic extraction	Low-Medium	Low or Medium
Meeting summary	Medium	Medium
Action item extraction	Medium	Medium
Content generation	Medium	Medium
Sentiment analysis	Medium	Medium
Multi-topic analysis	Medium-High	Medium or High
Research with tools	High	High
Strategic analysis	High	High
Complex reasoning	High	High

Cost vs Quality Tradeoffs

When to Use Low Tier

✅ Use Low when:

Task has clear, simple criteria
Speed is more important than nuance
Processing high volumes
Output is binary or simple

❌ Avoid Low when:

Task requires interpretation
Quality is critical
Input is ambiguous
Multi-step reasoning needed

When to Use Medium Tier

✅ Use Medium when:

Standard business tasks
Good quality matters
Balanced performance needed
Most common scenarios

❌ Avoid Medium when:

Task is trivially simple (use Low)
Task requires deep analysis (use High)

When to Use High Tier

✅ Use High when:

Complex analysis required
Stakes are high
Using AI Agent with tools
Nuanced judgment needed

❌ Avoid High when:

Task is straightforward (use Medium)
Speed is critical (use Medium/Low)
Cost is a major concern

Performance Comparison

Response Times (Typical)

Tier	Simple Task	Medium Task	Complex Task
Low	~2-5s	~5-10s	~10-15s
Medium	~3-8s	~8-15s	~15-30s
High	~5-15s	~15-45s	~30-90s

Times vary based on input size and task complexity

Quality Comparison

Simple Classification Task:

Low: 95%+ accuracy
Medium: 97%+ accuracy
High: 99%+ accuracy

Complex Analysis Task:

Low: May miss nuances
Medium: Good overall quality
High: Best quality, catches subtleties

Practical Examples

Example 1: Urgency Check

Task: Determine if meeting needs urgent follow-up

Recommendation: Low tier

yaml

model: low
return_type: boolean

Why: Binary decision with clear criteria.

Example 2: Meeting Summary

Task: Generate executive summary of meeting

Recommendation: Medium tier

yaml

model: medium
return_type: string

Why: Standard summarization, good quality needed.

Example 3: Deal Risk Analysis

Task: Identify risk factors and recommend actions

Recommendation: High tier

yaml

model: high
return_type: string

Why: Complex analysis requiring nuanced judgment.

Example 4: Prospect Research

Task: Research company using web search and documents

Recommendation: High tier

yaml

model: high
return_type: string
tools: [search, open, web_search]

Why: AI Agent with tools needs advanced reasoning.

Multi-Node Strategy

For complex workflows, use different tiers for different nodes:

text

[Load Meeting]
       │
       ├──▶ [AI Low: Is urgent?] ──▶ [If: urgent]
       │                                  │
       │                                  ├──▶ Yes ──▶ [AI High: Deep analysis]
       │                                  │
       │                                  └──▶ No ───▶ [AI Medium: Standard summary]
       │
       └──▶ [AI Low: Classify type] ──▶ [Route by type]

Benefits: