Tencent LifeSearch Benchmarks 12 Deep Research Agents on 203 Real-Life Tasks, GPT-5.6 Sol Tops at 63.4%

Tencent's LLM Department released LifeSearch, a 203-task bilingual benchmark for deep-research agents built from real-user query seeds across 7 life and productivity domains, with over 4,000 expert-written rubrics and 4,500 hours of human annotation. Across 12 frontier agents under a unified harness, the best model GPT-5.6 Sol scores 63.4%; a human-ranking study shows aggregate benchmark scores align with human judgment at Spearman's ρ=0.89.