Embedding-space map of LLMs — each model answers the same prompt battery; their response embeddings are concatenated and UMAP-projected to 2D so you can see which models cluster. - View it on GitHub