When should I use extraction vs. reasoning?
Use extraction when you need to identify and retrieve specific data points that are explicitly present in the document. Examples include extracting text strings, tables, lists, checkboxes, signatures, and barcodes. Extraction is suitable for tasks where the information is directly available and can be clearly defined by field names or prompts.
On the other hand, you should use reasoning when you need to generate results that are not explicitly found in the document but can be deduced, summarized, or calculated. Reasoning is useful for more complex tasks that require understanding context, performing calculations, or summarizing information. It involves using prompts to guide the model in generating the desired output based on the content of the document.
For example, use extraction for:
- Extracting a string of text or numbers, such as an address or account balance.
- Extracting tables or lists of items.
- Extracting checkboxes, signatures, or barcodes.
Use reasoning for:
- Generating summaries or calculations based on the document content.
- Deducing information that is not explicitly stated but can be inferred.
- Performing complex reasoning or computation queries.
For details, see our Building apps.