ALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model

ALTK-Evolve achieves top performance on the τ²-bench leaderboard for customer service tasks by leveraging learned memory mechanisms, outperforming models that rely on frontier architectures. The approach demonstrates that advanced memory integration can surpass reliance on large-scale model scaling in specialized domains.