代理建模旨在将LLM转变为能够通过规划,推理,工具使用,以及与外部环境的多轮交互来解决复杂任务的自主代理。我们提出 Orchard, 是一个用于可扩展代理建模的开源框架。其核心是 Orchard Env,，这是一种轻量级 Kubernetes 原生环境服务，为跨任务域, 代理利用, 和训练阶段的沙箱生命周期管理提供可重用原语。在 Orchard Env, 之上，我们构建了三个代理建模方案。 Orchard-SWE 的目标客户是软件工程代理。我们引入了信用分配监督微调和一系列 RL 信号: 平衡自适应推出(BAR)，用于稀疏奖励优化, 政策蒸馏(OPD) 和基于标题的过程奖励(RPR)，用于密集监督, 和历史经验蒸馏,，它将先前实验的推出压缩为用于推理时间重新排序的紧凑值模型。建立在 Qwen3.5-35B-A3B 主干, Orchard-SWE 上，基于 RPR 的 RL 达到 69.7%，在 SWE-bench Verified, 上通过价值模型重新排名达到 73.0%，在开源方法中树立了新的最先进水平，同时接近超过 10 倍的前沿系统。 Orchard-GUI 仅使用 0.4K 精炼轨迹和 2.2K 开放式任务, 训练 4B 视觉语言计算机使用代理，在 WebVoyager, Online-Mind2Web, 和 DeepShop, 上实现 68.4% 的平均成功率，使其成为最强大的开源模型，同时与专有系统保持竞争力。 Orchard-Claw 的目标是私人助理特工。仅使用 0.2K 综合任务, 进行训练，在 Claw-Eval 上达到 59.6% 通过@3，与更强的 ZeroClaw 安全带配对时达到 73.9%。总的来说,这些结果表明，轻量级,开放,与线束无关的环境层支持跨域可重用的代理数据,训练配方,和评估协议。
Agentic modeling aims to transform LLMs into autonomous agents capable of solving complex tasks through planning, reasoning, tool use, and multi-turn interaction with external environments. We present Orchard, an open-source framework for scalable agentic modeling. At its core is Orchard Env, a lightweight Kubernetes-native environment service that provides reusable primitives for sandbox lifecycle management across task domains, agent harnesses, and training stages. On top of Orchard Env, we build three agentic modeling recipes. Orchard-SWE targets software engineering agents. We introduce credit-assignment supervised fine-tuning and a progression of RL signals: Balanced Adaptive Rollout (BAR) for sparse-reward optimization, on-policy distillation (OPD) and rubric-based process reward (RPR) for dense supervision, and historical experience distillation, which compresses rollouts from prior experiments into a compact value model for inference-time reranking. Built on the Qwen3.5-35B-A3B backbone, Orchard-SWE reaches 69.7% with RPR-based RL and 73.0% with value-model reranking on SWE-bench Verified, setting a new state of the art among open-source methods while approaching frontier systems over 10x larger. Orchard-GUI trains a 4B vision-language computer-use agent using only 0.4K distilled trajectories and 2.2K open-ended tasks, achieving 68.4% average success across WebVoyager, Online-Mind2Web, and DeepShop, making it the strongest open-source model while remaining competitive with proprietary systems. Orchard-Claw targets personal assistant agents. Trained with only 0.2K synthetic tasks, it achieves 59.6% pass@3 on Claw-Eval and 73.9% when paired with the stronger ZeroClaw harness. Collectively, these results demonstrate that a lightweight, open, harness-agnostic environment layer enables reusable agentic data, training recipes, and evaluation protocols across domains.
科目: 人工智能 (cs.AI); 计算和语言 (cs.CL)
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)