比较不同标记格式在智能体系统中的效率，发现更紧凑的格式可节省27%令牌。
Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems
- 使用TOON和TRON等压缩格式替代JSON，减少智能体通信中的令牌开销。
- TRON最多减少27%令牌，准确率比JSON低14个百分点。
- TOON易因多轮解析失败导致错误传播，不适用于并行工具调用场景。
智能体系统中的大语言模型需处理工具模式与执行结果，并生成结构化工具调用。默认采用的JSON格式虽通用，但非为令牌效率设计，其结构元素带来显著令牌开销。近期提出如TOON（Token-Oriented Object Notation）和TRON（Token Reduced Object Notation）等更紧凑的替代格式，但仅在孤立的理解或生成任务中评估过。其在端到端智能体循环中的有效性仍未知。本文在四个智能体基准（BFCL、MCPToolBenchPP、MCP-Universe、StableToolBench）和五种开源大模型上评估TOON与TRON，独立测量输入压缩与输出压缩的效果。结果显示：TRON最多减少27%令牌，准确率比JSON低14个百分点；TOON最多减少18%令牌，准确率损失约9个百分点，但存在多轮解析失败时错误级联问题，且对多数模型导致并行工具调用输出崩溃。代码已公开于：https://github.com/lkutschka/notation-matters
原文摘要 · Abstract (English)
Large language models in Agentic AI systems consume tool schemas and execution results and emit tool invocations as structured data. The default language for that exchange, JSON, was designed for application-to-application interchange rather than token efficiency, so its structural elements impose substantial token overhead. Recent work proposes token-optimized alternatives such as TOON (Token-Oriented Object Notation) and TRON (Token Reduced Object Notation) as more compact replacements, but these formats have been evaluated only on isolated comprehension or generation tasks. Whether their token reductions hold inside end-to-end agentic loops therefore remains an open question. We evaluate TOON and TRON on four agentic benchmarks (BFCL, MCPToolBenchPP, MCP-Universe, StableToolBench) and five open-weight LLMs, decoupling input compression from output compression to measure comprehension and generation independently. TRON reduces tokens by up to 27% with accuracy within 14pp of the JSON baseline. TOON achieves up to 18% reduction at a similar 9pp accuracy cost, but additionally cascades on multi-turn parsing failures and collapses parallel tool-call output for most models. The code is available at: https://github.com/lkutschka/notation-matters
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。