ALTK-Evolve adds learned guidelines to a GPT-4.1 agent, lifting τ²-bench scores
Overview
IBM Research says ALTK-Evolve extracts behavioral guidelines from a GPT-4.1 agent's past training attempts and appends them to the system prompt, without updating model weights.
On held-out τ²-bench tasks, Pass^4 rises from 57.5% to 67.5% on retail with the full guideline set, and from 35.0% to 45.0% on airline with a compact core of guidelines. The post, published on Hugging Face, says the retail result is above every retail entry on the October 12, 2026 leaderboard snapshot. These figures come from IBM's own report; no independent evaluation is cited.
Written by AI from the articles below · updated Oct 20, 11:52 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Hugging Face BlogOfficialPickALTK-Evolve adds learned guidelines to a GPT-4.1 agent and tops τ²-bench retail
AIIBM Research says ALTK-Evolve extracts behavioral guidelines from a GPT-4.1 agent's training attempts and appends them to the system prompt, with no weight updates. On held-out τ²-bench retail tasks, Pass^4 rises from 57.5% to 67.5% with the full guideline set, and on airline from 35.0% to 45.0% with the compact core. The post reports that the retail result is above every retail entry on the October 12, 2026 leaderboard snapshot.
Heat Trend
- Current comparable heat
- 9
- Peak comparable heat
- 10Oct 20, 12:00 PM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.