"LLM benchmarks" are standardized tests or evaluation frameworks used to measure the performance, capabilities, and limitations of large language models (LLMs). These benchmarks typically include a variety of tasks—such as question answering, reasoning, summarization, code generation, and factual accuracy—that assess how well an LLM performs across different domains and skill sets. They help researchers and developers compare models, track progress, and identify areas for improvement.