The detail
At 13,300 tokens, a 10,000-word document uses about 10% of a 128K context window and under 7% of a 200K window, leaving ample room for instructions and a long response. Even multiple documents this size fit into the 1M-token windows now common on flagship models.
Sending it costs real but small money: at $1 per million input tokens, about $0.013 per request; at premium rates near $5 per million, about $0.067. The multiplier to watch is repetition — re-sending the document across a long conversation multiplies that cost by the number of turns, which is exactly what prompt caching (typically a 90% discount on cached input) is designed to fix.
For very long materials, chunking plus retrieval often beats whole-document prompting on both cost and answer quality, because models weight the most relevant context more heavily than bulk context.