LLM as a Council - Test Multiple Models to Get Best Output
CREATED BY
1 Template
98 Views
LAST UPDATED
November 23, 2025INTEGRATIONS USED
DESCRIPTION
LLM as a Council
Route your question through multiple models at once with an LLM as a judge function to pick the best one. Inspired by Karpathy's X Post.
When to Use
- Most helpful when deciding what model to use for specific task.
Integrations
- Chat interface to Ask AI nodes
Output
- Rates each reply and picks best one.
Additional builds: you can replace the interface node with an import data step to run evals on a whole CSV, or a slack / email reader, if you want the reply to meet you in a messaging interface.
HOW DO YOU SET THIS UP?
1.
Check the 4 models selected.
You can update the Ask AI nodes to other foundation model options. You can also add more if you want.
2.
Finalize the eval
This template uses LLM as a judge in a simple 1-10 response knowing the question and all of the replies. If you want to customize the criteria further for your use case you can update the prompt.
Hand-picked by the Gumloop team
Similar Templates
24 views
2 days ago
21 views
5 days ago
63 views
a week ago