How modern models handle million-token contexts and why retrieval still beats raw context length
15 views