CodeDance：一种用于可执行视觉推理的动态工具集成多模态大语言模型
CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning
将可执行代码作为视觉推理的通用求解器，通过动态组合工具（如检测、绘图）并渲染中间视觉产物，实现透明、可自检的推理过程；其强化学习训练中涌现的工具组合与跨任务迁移能力，为视觉语言模型的可解释推理与工具增强提供了新范式。
原文依据 CodeDance defines, composes, and executes code to orchestrate multiple tools, compute intermediate results, and render visual artifacts
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期发布的 o3 等模型凸显了人类般的“用图像思考”推理能力，它将工具使用与逐步验证相结合，然而大多数开源方法仍依赖纯文本链、僵化的视觉模式或单步流程，限制了复杂任务上的灵活性、可解释性和可迁移性。我们提出 CodeDance，探索将可执行代码作为视觉推理的通用求解器。与固定模式调用（例如仅预测边界框坐标）不同，CodeDance 定义、组合并执行代码来编排多个工具、计算中间结果并渲染视觉产物（如框、线、图表），以支持透明、可自检的推理。为引导这一过程，我们引入了一种平衡且自适应工具调用的奖励，它平衡探索与效率并缓解工具过度使用。有趣的是，除了原子监督所教授预期能力之外，我们在强化学习训练中经验性地观察到新颖的涌现行为：CodeDance 展现出新颖的工具调用、未见过的组合以及跨任务迁移。这些行为无需任务特定微调即可出现，表明了一种通用且可扩展的可执行视觉推理机制。在推理基准（如视觉搜索、数学、图表问答）上的大量实验表明，CodeDance 不仅持续优于模式驱动和纯文本基线，还超越了 GPT-4o 等闭源模型和更大的开源模型。
英文原文摘要 1499 字符
Recent releases such as o3 highlight human-like "thinking with images" reasoning that combines tool use with stepwise verification, yet most open-source approaches still rely on text-only chains, rigid visual schemas, or single-step pipelines, limiting flexibility, interpretability, and transferability on complex tasks. We introduce CodeDance, which explores executable code as a general solver for visual reasoning. Unlike fixed-schema calls (e.g., only predicting bounding-box coordinates), CodeDance defines, composes, and executes code to orchestrate multiple tools, compute intermediate results, and render visual artifacts (e.g., boxes, lines, plots) that support transparent, self-checkable reasoning. To guide this process, we introduce a reward for balanced and adaptive tool calling, which balances exploration with efficiency and mitigates tool overuse. Interestingly, beyond the expected capabilities taught by atomic supervision, we empirically observe novel emergent behaviors during RL training: CodeDance demonstrates novel tool invocations, unseen compositions, and cross-task transfer. These behaviors arise without task-specific fine-tuning, suggesting a general and scalable mechanism for executable visual reasoning. Extensive experiments across reasoning benchmarks (e.g., visual search, math, chart QA) show that CodeDance not only consistently outperforms schema-driven and text-only baselines, but also surpasses closed models such as GPT-4o and larger open-source models.