ButlerBot logo

ButlerBot

Alfred Models

A guide to AI models on Alfred and how each change Alfred's behaviour

Understanding Alfred's AI Models

Alfred leverages advanced AI models from OpenAI, Anthropic and Google, each carefully selected to provide optimal performance for different use cases. Below is a detailed breakdown of each model's capabilities and ideal use scenarios.

Summary

  • OpenAI

    • GPT-4: High Intelligence, Medium Costs

    • GPT-4-MINI: Low Intelligence, Low Costs

  • Anthropic

    • Claude-Sonnet: Very High Intelligence, High Costs

    • Claude-Haiku: Medium Intelligence, Somewhat Low Costs

  • Google

    • Gemini-Pro: High Intelligence, Medium-High Costs

    • Gemini-Flash: Somewhat Low Intelligence, Low Costs

Our recommendation:

  • Claude-Sonnet for high intelligence tasks (most intelligent model)

  • Gemini-Pro for higher-medium intelligence tasks

  • Claude-Haiku for medium intelligence tasks (perfect middleground)

  • GPT-4 for general and balanced performance

  • Gemini-Flash for low-cost, high-volume tasks (cheapest model)

Note: a cheaper model means that you can use more of that model before running out of daily/weekly usage for that model

Model Overview

OpenAI Models

GPT-4 (GPT-4.1)

  • Strong Suits:

    • Comprehensive multimodal processing

    • Superior performance in non-English languages

  • Best For:

    • Multilingual interactions

    • Professional applications

    • Mid-volume interactions

GPT-4-MINI (GPT-4.1-mini)

  • Strong Suits:

    • Optimized for speed and efficiency

    • Cost-efficient operations

  • Best For:

    • Quick responses

    • Routine tasks with low analytical requirements

    • High-volume interactions

NOTE: Gemini-Flash is more intelligent, faster and lower cost than GPT-4-MINI

Anthropic Models

Claude-Sonnet (Claude 3.7 Sonnet)

  • Strong Suits:

    • Advanced coding capabilities

    • Persistent problem solving

    • Enhanced reasoning and analysis

  • Best For:

    • Programming tasks

    • Technical queries with high analytical requirements

    • Highly Intelligent interactions

Claude-Haiku (Claude 3.5 Haiku)

  • Strong Suits:

    • Optimized for speed and efficiency

    • Somewhat Cost-efficient operations

  • Best For:

    • Quick responses

    • Routine tasks with medium analytical requirements

    • High-volume interactions

Google Gemini Models

Gemini-Pro (Gemini 2.5 Pro)

  • Strong Suits:

    • Effecient input processing (good for processing large inputs)

    • Good cost to performance ratio

  • Best For:

    • Complex analytical tasks

    • Intelligent interactions at a reasonable cost

Gemini-Flash (Gemini 2.0 Flash)

  • Strong Suits:

    • Optimized for speed and efficiency with good intelligence

    • Lazy model, will not over do tasks

    • Cost-efficient operations

  • Best For:

    • Quick responses

    • Routine tasks with low analytical requirements

    • High-volume interactions

Choosing Your Model

  • For maximum capabilities: Claude-Sonnet

  • For advanced reasoning and thinking: Gemini-Pro

  • For smart, low-cost, speed-critical tasks: Claude-Haiku

  • For balanced performance and cost: GPT-4

  • For extreme-low-cost, speed-critical tasks: Gemini-Flash

Our Recommendation

During our own internal testing, we've found that Claude-Sonnet consitently outperforms other models in terms of accuracy, intelligence and our most important metric - engagement. However, for cost-sensitive applications, GPT-4, Gemini-Pro and Claude-Haiku are a great alternatives. Finally, for high-volume and surface-level operations, Gemini-Flash is the optimal model.

Performance Matrix

Aspect GPT-4 GPT-4-MINI Claude-Sonnet Claude-Haiku Gemini-Pro Gemini-Flash
Speed Good Excellent Good Excellent Good Best
Accuracy Excellent Medium Best Good Excellent Good
Features Full Basic Full Full Full Full
Cost-Efficiency Standard Excellent Standard High Standard Best

Based on internal testing and experiences