[TiDB docs, vector-search-data-types]: the VECTOR type enforces its dimension at insert time. Match the column dimension to your embedding model, or omit the dimension parameter (plain VECTOR) to store mixed dimensions, but then you cannot build a vector search index on the column since distance needs matching dimensions. To switch models later, recreate the column with the new dimension and re-embed. Invalid literal syntax like '[5, ]' gives Invalid vector text.

Context: Inserts fail with ERROR 1105 (HY000): vector has 2 dimensions, does not fit VECTOR(3).