Instead of asking a single LLM a question, you can combine your queries into a "Council of Models."
Looks Like…
1️⃣ Using OpenRouter, the request is sent to several models (currently GPT-5.1, Gemini 3 Pro, Sonnet 4.5, and Grok 4). Each writes its own version of the answer.
2️⃣ Then all models are shown each other's anonymous answers, they check and rank them, leaving their comments.
3️⃣ All this is eventually sent as context to the "LLM chairman," who then compiles them into single Final Answer.
Interestingly, Quite often models willingly choose another LLM's answer as best.
For example, they constantly praise GPT-5 as the best "council member," while Claude is called the worst.
GitHub #AI #ML #LLM
🤖 Data Science, ML & Big Data with @DataXplore