AgentBench: Evaluating LLMs as Agents

AgentBench is a multi-dimensional benchmark that consists of 8 distinct environments for evaluating LLMs as autonomous agents in interactive tasks. It assesses reasoning and decision-making abilities in multi-turn open-ended generation, and tests over 25 LLMs including APIs and open-source models.

Website Directory import
0 shouts Added Sep 23, 2026 huggingface.co
huggingface.co/papers/2308.03688
AgentBench: Evaluating LLMs as Agents website preview

Links

Tags