Verified70% confidenceFactExact time
A benchmark is a fixed set of questions used to score and horizontally compare models, similar to a mock exam, and is used to rank models.
4
Sources
70%
Confidence
Long-term
Relevance
9/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBEIR Benchmark是一套跨领域的检索评估基准，包含多个数据集和标准指标（nDCG、Recall），是系统对比检索策略的首选。78% similarUnverified评测游戏（Benchmark Gaming）指AI公司通过选择性使用特定基准测试来最大化模型表现分数的行为，包括数据污染和针对特定评测集过度优化73% similarUnverifiedcriterion is typically used with the cargo bench command, defining benchmarks as independent test targets.73% similarUnverifiedAgentic benchmarks are fundamentally different from traditional single-turn question-answering evaluations: the former simulate a complete workflow, requiring the model to autonomously decompose tasks, call tools, maintain multi-step intermediate state, and deliver verifiable results.72% similarUnverifiedBenchmark saturation refers to the loss of discriminative power of a benchmark when multiple models approach perfect scores on it, prompting the community to introduce new evaluation standards such as GPQA and SWE-bench.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/943510API
curl https://kongchang.com/api/v1/knowledge/claims/943510MCP
get_claim(id=943510)