Options A, B, and C are correct. Snowflake explicitly recommends splitting text in a search column into chunks of no more than 512 tokens for best search results with Cortex Search. Research indicates that smaller chunk sizes typically result in higher retrieval precision for a given query and improved downstream LLM response quality. This practice is recommended even when longer-context embedding models, such as

with an 8000 token context window, are available, because smaller chunks provide more precise retrieval and more relevant context for the LLM. Option D is incorrect; the sources do not mention that SPLIT_TEXT_RECURSIVE_CHARACTER automatically enriches chunks with metadata. This would typically require additional data processing steps. Option E is incorrect; the primary goal of text splitting is to optimize retrieval and LLM response quality, not to balance input and output token counts for cost reasons. While token counts influence costs, the 512-token recommendation is driven by quality considerations.