​
Onyx’s Approach to LLM Providers

Onyx is designed to be model-agnostic, offering you the flexibility to choose the Language Model (LLM) that best suits your needs. This approach ensures that you’re not locked into a single provider and can leverage the strengths of different models for various tasks.

​
Model Overview

Onyx offers integration with several popular LLM providers, with a focus on models from OpenAI and Anthropic. Here’s an overview of key models:

​
OpenAI Models

​
GPT-3.5-Turbo

  • Strengths: High speed, good quality for general tasks
  • Best for: Quick queries, general information retrieval
  • Knowledge cutoff: September 2021

​
GPT-4

  • Strengths: High-quality responses, strong reasoning capabilities
  • Best for: Complex analysis, creative tasks, code generation
  • Knowledge cutoff: April 2023
  • Note: Capable of image analysis

Here’s the updated information on Anthropic models, including Claude 3.5 Sonnet and a recommendation:

​
Anthropic Models

​
Claude-3 Opus

  • Strengths: Exceptional performance in reasoning and analysis tasks
  • Best for: Complex problem-solving, detailed explanations
  • Note: Offers high accuracy but may have slower response times

​
Claude-3 Sonnet

  • Strengths: Balance of performance and speed
  • Best for: General-purpose tasks requiring good quality and reasonable speed
  • Note: Good all-around performer for most use cases

​
Claude 3.5 Sonnet

  • Strengths: Enhanced capabilities over Claude-3 Sonnet
  • Best for: Advanced general-purpose tasks with improved performance
  • Note: Recommended for most use cases due to its superior balance of capabilities

​
Claude-3 Haiku

  • Strengths: Fast responses, efficient for simpler tasks
  • Best for: Quick queries, real-time applications
  • Note: Sacrifices some complexity for speed

For most use cases, Claude 3.5 Sonnet is recommended as it offers an excellent balance of advanced capabilities, performance, and speed, making it suitable for a wide range of applications.

​
Custom Providers

Onyx allows you to add custom providers by integrating any model from the LiteLLM providers list. This flexibility enables you to use specialized or proprietary models that best fit your organization’s needs.

​
Choosing the Right Model

Consider these factors when selecting a model:

  1. Task Complexity: More complex tasks benefit from advanced models like GPT-4 or Claude-3 Opus.
  2. Response Speed: For quick responses, consider faster models like GPT-3.5-Turbo or Claude-3 Haiku.
  3. Cost Considerations: More advanced models typically have higher usage costs.
  4. Data Privacy: For strict data policies, consider open-source models like Llama 2 that can be self-hosted.
  5. Specific Strengths: Some models excel in particular areas (e.g., coding, analysis, creativity).

We recommend testing different models with your typical queries to determine which performs best for your specific use cases.

​
Leveraging Model Flexibility

  1. Experiment: Try different models for the same task to compare results.
  2. Monitor Performance: Track which models perform best for different types of queries.
  3. Stay Updated: Regularly check for updates and new model releases.
  4. Custom Integration: Explore the option of integrating specialized models from the LiteLLM providers list for unique use cases.
Creating AssistantsEmbedding Models