IBM Research

IBM Research says its ALTK-Evolve method took a stock ReAct agent running GPT-4.1 to 67.5% Pass^4 on τ²-bench retail, above every retail entry in its October 12, 2026 leaderboard snapshot. The method appends learned guidelines to the system prompt and updates no model weights.

IBM Research introduces ALTK-Evolve, an agentic memory system that matches or beats ACE's accuracy on the AppWorld benchmark while cutting inference token costs by up to 86% through selective guideline retrieval instead of full-playbook injection.