Search papers, labs, and topics across Lattice.
2
0
5
Current language agents are still far from matching human expert performance when faced with real-world professional tasks requiring complex reasoning, authoritative source retrieval, and domain-specific knowledge, as revealed by the new \$OneMillion-Bench benchmark.
Ditch the deep thought: this new agentic search framework slashes reasoning steps by 70% while boosting accuracy by prioritizing parallel evidence gathering.
