Estimates based on input token pricing only. Actual savings depend on conversation length, document corpus size, and query patterns. Conversation history estimated at ~600 tokens per message. Memory retrieval modeled from observed Exabase extraction ratios.
