Three agents, one question
Agents answered the same 19 questions about Vulcan, Alberta with three different sets of tools: today's page, the page and its edit history, or Agent Memory holding that history. Each question ran three times per agent, and a separate model scored every answer from 0 to 2 against a written answer key without knowing which agent wrote it.
Accuracy and cost
All 19 questions
Each cell is the mean score of three runs, from 0 to 2. Select a question to read the answers.