ALTK-Evolve Learning Memory Tops τ²-bench Customer Service Leaderboard
Understand the story
On October 20, 2026, IBM Research published the ALTK-Evolve method on the Hugging Face Blog. The method distills behavioral guidelines from an agent's own trajectories and, simply by appending these guidelines to the system prompt, achieves learning-type memory. A GPT-4.1-based ReAct customer service agent reached Pass^4 67.5% on τ²-bench retail, topping the τ²-bench customer service leaderboard. This is currently the only report on the event, with no further developments.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- Hugging Face BlogSelectedALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model
IBM Research released the ALTK-Evolve method, which simply appends behavioral guidelines distilled from the agent's own trajectories to the system prompt, enabling a GPT-4.1-based ReAct customer service agent to reach Pass^4 67.5% on τ²-bench retail.
Story heat trend
Not enough continuous observations to show a trend yet.