New architectures and benchmarks advance large language model agent reasoning and evaluation

Recent research introduces innovative architectures and benchmarks to improve large language model (LLM) agents' reasoning efficiency and reliability.