用无训练的协作框架提升大模型编程能力，效果因模型而异
Zero-Shot Self-Orchestration with Ledger-Based Control for Improved LLM Coding Performance
- 引入管理-执行架构，通过共享文件系统实现零样本协同
- 部分模型单次通过下准确率提升超23%，成本仅为大模型的五分之一
- 适合资源有限但需稳定性能的开发者，尤其在禁用推理时效果更佳
多智能体大模型系统常被认为优于单模型，但证据混杂：管道变化同时影响令牌预算、工具调用和提示词，难以分辨真正贡献。我们研究在共享文件系统工作区上引入管理-执行架构的影响，不进行训练或特定基准调优，与单次调用的同一模型对比。覆盖九个模型——五个开源（9B至~2.8T参数）和四个前沿闭源模型，在最新的100道硬级LiveCodeBench问题上，该架构带来的收益真实但有条件：对部分模型显著（Qwen3.8-27B +23.4，GPT-5.6-Luna +10.6，GPT-5.6-Terra +8.0，均在五次配对通过中；Kimi-K3 +30.4，Minimax-M3 +11.0，五次配对通过且关闭推理，均p < 10^-4；单次通过128k上下文时+42和+12）；对其他模型则无效甚至为负（Qwen3.6-35B 关闭推理时-1至-9）。启用管理器后，Opus-5 在单次通过中达91%最高分。运行管理器使令牌开销约增至三倍，但换来的准确性远比升级模型更划算：GPT-5.6-Terra 使用管理器几乎达到Fable 5的单次调用准确率（85.0对87.4，p=0.59），成本仅为五分之一（$11.71对$61.11/100题，p<10^-4），Qwen-27B 仅需$51.75且可自托管。转录分析揭示两种主要机制：上下文管理（短调用与共享笔记减少截断）与问题分解。大模型开启推理时提升小，而小模型开启推理或部分模型关闭推理时提升更大。
原文摘要 · Abstract (English)
Multi-agent large language model systems are widely reported to beat single-model baselines, but the evidence is mixed, and comparisons are usually confounded: pipelines change token budgets, tool calls, and prompts simultaneously, so an aggregate gain rarely reveals what actually helped. We investigate the effect of introducing the manager-worker scaffold over a shared filesystem workspace, with no training and no per-benchmark tuning, measured against the same model answering in a single pass. Across nine models -- five open-weight, spanning 9B to ~2.8T parameters, and four frontier closed models -- on the 100 latest hard LiveCodeBench problems, the scaffold's benefit is real but conditional: large and statistically significant for some (Qwen3.8-27B +23.4, GPT-5.6-Luna +10.6 and GPT-5.6-Terra +8.0, each over five paired passes; Kimi-K3 +30.4 and Minimax-M3 +11.0 over five paired passes with reasoning off, both at $p < 10^{-4}$, and +42 and +12 in a single pass at a 128k cap) and null or negative for others (Qwen3.6-35B -1 to -9 with reasoning off). With the manager, Opus-5 achieves the highest score in the study at 91% in one pass. Running a manager roughly triples the token bill, but it buys accuracy more cheaply than moving to a larger model does: GPT-5.6-Terra with a manager nearly matches Fable 5's single-call accuracy (85.0 against 87.4, $p = 0.59$) at a fifth of the price (\$11.71 against \$61.11 per 100-problem pass, $p < 10^{-4}$), and the Qwen-27B arm does it for \$51.75 on weights anyone can self-host. Our transcript analysis finds several mechanisms behind the gains, of which two recur: context management, in which short worker calls and shared notes organize state and reduce truncation, and problem decomposition. Improvements are modest for large models with reasoning enabled, but larger for some models with reasoning disabled and for smaller models with reasoning enabled.
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。