Compar:IA is an open-source LLM arena created by the French government, where users blindly compare AI models in real conversations. The resulting data is continuously published as three open-source datasets (conversations, votes, reactions) available for research and model training.

The map below shows 2,000 conversations (sampled from 25,000 total), positioned by semantic similarity and colored by topic. Search to highlight matching points, or click any point to see its details.

Contact: contact@comparia.beta.gouv.fr