Choosing Between Reasoning and General Purpose AI Models

Jylo offers both reasoning and general purpose AI models, each designed for different types of tasks. Understanding when to use each model type will help you optimise analysis quality, speed, and cost.

This article explains the key differences between model types and provides practical guidance on selecting the right model for your task.

Quick Decision Guide

Note: GPT-5 is currently the only routing model available. It automatically determines whether to use reasoning or general purpose capabilities for each prompt. Its best used as a reasoning model thanks to its similar level of intelligence and processing speed. 

Key Differences

General Purpose Models

• Conversational and user-friendly
• Fast processing speeds
• Suitable for standard tasks
• General understanding of concepts

Available General Purpose Models:

• GPT-4.1
• GPT-4o
• LLama 3.3
• Mistral Large 2411
• Phi 3 Medium
• Phi 4

Reasoning Models

• Concise, focused outputs
• Deeper domain-specific understanding
• Superior nuance detection
• Longer processing times

Available Reasoning Models:

• GPT-5 
• o3 (highest reasoning capability)
• o3-mini (recommended for balanced speed and intelligence)
• Deepseek R1

Available Routing Model:

Any model on Jylo preceded by an "o" is a reasoning model. "Mini" versions are faster than their sibling as they have been reduced to just the essential "parameters" enabling them to retain almost all of their intelligence whilst reducing processing time thanks to their smaller architecture. 

Recommended Starting Point: For most tasks requiring reasoning, o3-mini offers an excellent balance between analytical depth and processing speed. GPT5 is slightly more intelligent but is also marginally slower to process text.

Comparing Model Outputs

To illustrate the practical difference, here's how each model type responds to the same prompt.

Prompt: "Analyse this share purchase agreement."

General Purpose Model Output:

"This is a share purchase agreement between a buyer and seller for the transfer of company shares. The document includes standard sections such as purchase price, warranties, completion mechanics, and post-completion obligations. The purchase price is £500,000 with completion scheduled for 30 days after signing. I can provide more detail on any specific section if needed."

Reasoning Model Output:

"This share purchase agreement contains several noteworthy provisions. The locked box mechanism shifts economic risk to the buyer from the accounts date, but the leakage provisions are narrowly drafted and may not capture all value transfers. The warranty de minimis threshold of £5,000 is unusually low, potentially exposing the seller to minor claims. The tax covenant survives for seven years, exceeding the standard HMRC enquiry window, with an aggressive 100% liability cap. The three-year UK-wide non-compete clause may face enforceability challenges given the company's regional South East operations."

The Difference: General purpose models provide structural summaries, whilst reasoning models identify specific risks and commercial implications without explicit instruction.

When to Use Each Model Type

General Purpose Models

• Standard, familiar documents (e.g., typical NDAs)
• High-volume processing in a flow where speed matters
• Quick analysis of routine tasks
• Cost efficiency is a priority

Reasoning Models

• Complex or unusual documents with potential nuanced provisions
• Situations where you lack detailed knowledge of the document's content
• When concise, focused outputs are needed without conversational content
• Documents where the end goal or potential issues are unclear

Selecting Models for Flows and Playbooks

When building or running Flows, model selection has a compounding effect on both time and cost, as each prompt is executed against every document in your dataset.

Use General Purpose Models When:

• Extracting straightforward information (e.g., party names, dates, contract values)
• Using simple Yes/No prompts or List of Values prompts
• Reviewing many documents with standard prompts
• Processing high-volume datasets where speed matters

Example Scenario: Extracting landlord names, lease terms, and break clause presence from 500 lease agreements. General purpose models will handle these factual extractions quickly and cost-effectively.

Use Reasoning Models When:

• Prompts require nuanced interpretation or domain expertise
• Asking the AI to identify risks, assess compliance, or evaluate quality
• Working with Generative Output prompts that need sophisticated analysis
• The prompt involves complex conditional logic or multi-step reasoning

Example Scenario: Analysing share purchase agreements to identify unusual warranty provisions, assess the adequacy of indemnity protections, and flag commercially aggressive terms. Reasoning models will provide deeper insights worth the additional processing time.

Mixed Approach for Complex Playbooks:

Consider using different models for different prompts within the same Playbook:

• General purpose models for factual extraction prompts (dates, values, names)
• Reasoning models for analytical prompts (risk assessment, compliance evaluation)

This hybrid approach optimises both speed and analytical quality.

Understanding the Compounding Effect:

A Playbook with 20 prompts running on 100 documents means 2,000 individual AI requests. If each reasoning model prompt takes 30 seconds versus 5 seconds for a general purpose model:

• All reasoning models: ~16.5 hours to complete
• All general purpose models: ~2.8 hours to complete
• Hybrid approach (5 reasoning, 15 general): ~5.5 hours to complete

Choose models strategically based on what each individual prompt needs to accomplish.

Model Recommendations

For the most up-to-date model recommendations and specific model capabilities, visit our article How do I choose the right AI model in Jylo?

Key Principles:

• Use mini reasoning models (e.g., o3-mini) for deep analysis without excessive time or cost
• Use general purpose models for speed and standard tasks
• Match model capability to task complexity

Important Reminder

Whilst AI models are powerful tools, they don't replace professional expertise gained through years of experience. You remain responsible for ensuring AI outputs meet your professional standards. Always verify critical analysis and use AI as a tool to augment, not replace, your judgement.

Need Help?

Contact our support team at support@jylo.ai.