LLM Models - supercog-ai/community GitHub Wiki

Supercog supports the following families of LLM models:

• GPT models from OpenAI
• Claude models from Anthropic
• LLama3x models from Facebook, via Groq

Open the Settings page to set your OpenAI, Anthropic, or Groq API keys.

To change the active model for your agent, open the Settings section in the lower-left of the agent editor.

The following describes our recommendations for choosing your LLM:

Best Overall - Claude Sonnet 3.5

At present Claude offers the best combination of advanced reasoning capabilities and reasonable pricing. Using Claude is recommended when experimenting and developing new agents.

Best Value - GPT4o-mini

The default LLM selected when you create a new agent is GPT4o-mini. This models offers good reasoning capability, reliable results, and costs a fraction of the other leading models (0.15 cents/1k tokens vs. 3 cents/1k tokens for Claude 3.5).

Advanced Reasoning - GPT-4o

The GP-4o model is quite expensive (10 cents/1k tokens) but it offers very advanced reasoning and instruction following. We recommend trying advanced agents with Claude 3.5 or GPT-mini first, and then resorting to GPT4 if you don't get the desired behavior.