Shopping Reasoning Bench tests multi-turn retail assistants

Shopping Reasoning Bench targets a practical weakness in AI shopping assistants: real purchases require messy, multi-turn trade-offs.

The benchmark includes expert-authored shopping missions with importance-weighted rubrics, covering preferences, budgets, product comparisons, and follow-up clarification.

That makes it more realistic than single-turn factual QA for commerce systems, where a good answer often depends on balancing subjective user priorities rather than finding one objectively correct item.