Choosing an embedding model for short, messy, multilingual text
Five models, two engines and a brand-new release, all tested on my own text, on my own CPU. The model with the best score was not the one I picked, and the engine mattered more than I expected.