101 languages, one embedding space

101, all in a single embedding space. The lang filter accepts the full ISO set; the 30 highest-volume languages have their own reference pages.

No. Semantic and hybrid modes cross languages natively, because texts with the same meaning land near each other in the shared space regardless of language. Translating the query would search for the translation rather than the meaning.

No — keyword mode matches tokens, and an English token does not appear in a German document. Use semantic or hybrid mode for cross-language retrieval. Hybrid, the default, gives you lexical precision within the query language and semantic reach across all of them.

As character bigrams, because those scripts do not delimit words with spaces. 量子力学 indexes as 量子, 子力 and 力学, with the same rule applied to queries so the two always agree. It needs no language-specific segmentation model.

No, and it would be misleading to imply otherwise — English is by far the largest slice of the corpus. Set facets=true to see the language distribution of a result set before you answer, rather than discovering thin coverage from a user complaint.