Brief, mostly facts
IBM's ALTK-Evolve lifts GPT-4.1 to 67.5% on tau2-bench retail
Appends LLM-distilled guidelines to the system prompt, with no weight updates. The result is reported above all retail entries on the Oct 12 snapshot.
What the research found
IBM Research's ALTK-Evolve gets GPT-4.1 with a stock ReAct agent to 67.5% Pass^4 on tau2-bench retail by appending LLM-distilled guidelines to the system prompt, with no weight updates, reportedly above all retail entries on the October 12, 2026 snapshot.
Claims and how well the evidence supports them
- GPT-4.1 ReAct agent reaches 67.5% Pass^4 on tau2-bench retail with ALTK-Evolve 98%
- Learned guidelines are appended to the system prompt with no weight updates 99%
- Score is above every retail entry on the October 12, 2026 leaderboard snapshot 98%
Evidence
- huggingface.co https://huggingface.co/blog/ibm-research/altk-evolve-bc
Actionability rated low.