Introducing Muse Code and Muse Spark 1.2隆重推出 Muse Code 与 Muse Spark 1.2

August 5, 20262026年8月5日4 minute read4分钟阅读·4分钟阅读

Muse Code and Muse Spark 1.2 launch artwork.

We're excited to release Muse Code (beta), a terminal coding agent powered by Muse Spark 1.2, our newest model. This marks our next step toward the frontier, with larger and much more capable models on the way.我们很高兴发布 Muse Code (测试版)，这是一款由我们最新模型 Muse Spark 1.2 驱动的终端编码智能体。这是我们迈向技术前沿的又一步，未来还将有更强大、能力更卓越的模型问世。

Install Muse Code on macOS or Linux:在 macOS 或 Linux 上安装 Muse Code：

Muse Code takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results. It can coordinate multiple persistent subagents for each task, solving difficult problems faster, more accurately, and with less intervention.Muse Code 能够处理大型代码库中复杂的软件工程任务：规划变更、编写代码并验证结果。它能为每项任务协调多个持久化的子智能体，从而以更快的速度、更高的精度解决难题，并减少人工干预。

Muse CodeMuse Code

Async Background Agents异步后台智能体

Muse Code operates with a simple agent loop plus a set of async background agents to enhance the main agent's capability. These specialized background agents remain active throughout each session, rather than being spawned for individual tasks, helping avoid redundant information gathering. They carry out next steps and choose when to communicate back to the main agent. Their persistence reduces latency and the need for steering on difficult, multi-step tasks.Muse Code 采用简单的智能体循环，并结合一组异步后台智能体来增强主智能体的能力。这些专业的后台智能体在整个会话期间保持活跃，而不是针对单个任务临时生成，这有助于避免重复收集信息。它们会执行后续步骤，并选择何时与主智能体进行通信。这种持久性降低了延迟，并减少了在处理复杂的多步骤任务时进行人工引导的需求。

Runtime Design运行时设计

Muse Code uses a local event log in which every model call, tool run, approval, and edit is appended. This single source of truth makes the runtime replay-exact and restart-safe: after a crash, the agent can resume precisely where it stopped. That ability lets Muse Code take on long-running tasks without being derailed by failures.Muse Code 使用本地事件日志，其中会追加记录每一次模型调用、工具运行、批准和编辑操作。这一单一事实来源使得运行时能够实现精确重放和重启安全：即使发生崩溃，智能体也能从中断处准确恢复。这种能力使 Muse Code 能够承担长期运行的任务，而不会因故障而中断。

Bundled Skills内置技能

Muse Code ships with several default skills. /plan turns a task into an approval-gated plan, /grill stress-tests that plan until it holds up, and /goal works toward successful completion of the specified objective.Muse Code 附带了几项默认技能。/plan 可将任务转化为带有批准门控的计划；/grill 可对该计划进行压力测试，直到其稳健可行；/goal 则致力于顺利完成设定的目标。

The user inputs a fly-through video of a home into the terminal as an mp4 file. Muse Code interprets the video and produces a visually rich vacation home marketing and booking page.用户将一段房屋漫游视频作为 mp4 文件输入到终端。Muse Code 会解读该视频，并生成一个视觉丰富、包含预订功能的度假屋营销页面。

Muse Spark 1.2Muse Spark 1.2

Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. In Muse Spark 1.2, we significantly scaled up training compute on coding tasks while expanding training environment diversity. The model also maintains its strength in other key areas like general agents.Muse Spark 1.2 是 Muse Spark 1.1 的编码专项更新版本，在代码生成、复杂调试、代码库理解以及端到端开发者工作流方面均有改进。在 Muse Spark 1.2 中，我们显著提升了编码任务的训练算力，并扩大了训练环境的多样性。该模型在通用智能体等其他关键领域也保持了其优势。

Bar chart comparing Terminal-Bench 2.1 scores for Muse Spark 1.2 and other coding models.

Bar chart comparing DeepSWE 1.1 scores for Muse Spark 1.2 and other coding models.

Bar chart comparing Meta Internal Coding Bench scores for Muse Spark 1.2 and other coding models.

For more details about our evaluations, see our report.有关我们评估的更多详细信息，请参阅我们的报告。

Co-Training With Muse Code与 Muse Code 联合训练

We co-trained Muse Spark 1.2 with Muse Code to ensure the model exhibits its best performance and coding usability when paired together. The training included rejection sampled harness trajectories and recipe optimizations for goals, compaction, and subagents, alongside the integration of the Muse Code toolset to maximize harness compatibility.我们对 Muse Spark 1.2 与 Muse Code 进行了联合训练，以确保模型在两者配合使用时展现出最佳性能和编码易用性。训练过程包括拒绝采样后的工具链轨迹、针对目标、压缩和子智能体的方案优化，以及 Muse Code 工具集的集成，从而最大限度地提高工具链的兼容性。

Long-Horizon长周期任务

Muse Spark 1.2 was extensively trained on long-horizon coding tasks, including whole-repository generation, large end-to-end projects, and auto-research. It leverages planning to sequence work, goal conditioning to maintain direction, and context compaction to retain the knowledge needed to sustain progress.Muse Spark 1.2 针对长周期编码任务进行了广泛训练，包括全代码库生成、大型端到端项目和自动研究。它利用规划来安排工作顺序，利用目标条件来保持方向，并利用上下文压缩来保留维持进展所需的知识。

Self-Improvement自我提升

We also used Muse Spark 1.1 to generate challenging coding environments and instruction-following templates. The model then graded candidate solutions on how well they satisfied those requirements, producing a scalable training dataset for Muse Spark 1.2. This self-improvement loop helped Muse Spark 1.2 follow complex instructions more precisely than its predecessor.我们还使用 Muse Spark 1.1 生成了具有挑战性的编码环境和指令遵循模板。模型随后会对候选解决方案进行评分，评估其满足这些要求的程度，从而为 Muse Spark 1.2 生成可扩展的训练数据集。这种自我提升循环帮助 Muse Spark 1.2 比其前代产品更精确地遵循复杂指令。

Case Study: Kernel Optimization案例研究：内核优化

We tested the model's ability to iteratively optimize GPU kernels over 1,000+ tool calls (up to 24 hours). Leveraging Muse Code's agentic coding environment, the model writes, compiles, profiles, and progressively improves kernel performance relative to a provided baseline implementation. We benchmarked on KDA and MLA kernels for NVIDIA Hopper GPUs. The agent continues to achieve substantial improvements over the provided baseline implementation.我们测试了该模型在超过 1,000 次工具调用（长达 24 小时）中迭代优化 GPU 内核的能力。利用 Muse Code 的智能体编码环境，模型可以编写、编译、分析并逐步提升内核性能，使其优于提供的基准实现。我们对 NVIDIA Hopper GPU 上的 KDA 和 MLA 内核进行了基准测试。该智能体在提供的基准实现基础上持续实现了显著的性能提升。

Chart comparing KDA kernel speedup against the baseline over cumulative tool calls for Muse Spark 1.2 and other models.

The baseline is the FLA Triton implementation of KDA. Models were prohibited from importing third-party kernel libraries such as FLA directly; instead, they had to apply specialized kernel-optimization knowledge to implement the algorithm in Triton, rather than wrap existing implementations. Muse Spark 1.2 paired a chunk-parallel preparation kernel with a sequential inter-chunk scan, combining standard fusion and tiling with KDA-specific optimizations such as re-centering the gated cumulative decay at the chunk midpoint.基准测试采用的是 KDA 的 FLA Triton 实现。模型被禁止直接导入第三方内核库（如 FLA）；相反，它们必须应用专业的内核优化知识，在 Triton 中实现算法，而不是简单地封装现有实现。Muse Spark 1.2 将块并行准备内核与顺序块间扫描相结合，将标准融合与分块技术与 KDA 特有的优化（如在块中点重新调整门控累积衰减）相结合。

Availability可用性

Muse Spark 1.2 is available today in Muse Code and in Meta Model API with expanded global access. We have a lot on the horizon, including new harness features and more powerful models. We can’t wait to see what you build!Muse Spark 1.2 现已在 Muse Code 和 Meta Model API 中提供，并扩大了全球访问权限。我们未来还有许多计划，包括新的工具链功能和更强大的模型。我们迫不及待地想看到你们创造的作品！