Muse Spark 1.2 vs. DeepSWE

Muse Spark 1.2 is a coding-focused AI model from Meta designed to support advanced software engineering tasks through Muse Code and the Meta Model API. The model builds on Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. Muse Spark 1.2 powers Muse Code, a terminal coding agent that can plan repository changes, write code, validate outputs, and work across large codebases. Muse Code uses persistent async background agents that stay active throughout a session to reduce redundant information gathering and support difficult multi-step work. The runtime uses a local event log where model calls, tool runs, approvals, and edits are appended, making sessions replay-exact and restart-safe. Muse Spark 1.2 was co-trained with Muse Code so the model can take advantage of its toolset, harness workflows, goals, compaction, and subagent architecture. Meta significantly scaled training compute on coding tasks and expanded training environment diversity to improve the model’s engineering capabilities. The model was also trained on long-horizon coding tasks, including whole-repository generation, large end-to-end projects, auto-research, and extended iterative work. Its training approach uses planning, goal conditioning, context compaction, rejection-sampled harness trajectories, and self-improvement data generated with Muse Spark 1.1. Meta also tested Muse Spark 1.2 on long-running GPU kernel optimization workflows where the model wrote, compiled, profiled, and improved Triton kernels over many tool calls. By combining coding-focused training, agentic runtime integration, persistent subagents, long-horizon reasoning, replay-safe execution, and API availability, Muse Spark 1.2 helps developers and AI agents complete complex software engineering work with less intervention.

DeepSWE represents a groundbreaking advancement in open-source coding agents, harnessing the Qwen3-32B foundation model trained exclusively through reinforcement learning (RL) without the aid of supervised fine-tuning or proprietary model distillation. Developed using rLLM, which is Agentica's open-source RL framework tailored for language-driven agents, DeepSWE functions effectively within a simulated development environment provided by the R2E-Gym framework. This setup equips it with a range of tools, such as a file editor, search functions, shell execution, and submission capabilities, allowing the agent to adeptly navigate extensive codebases, modify multiple files, compile code, execute tests, and iteratively generate patches or fulfill intricate engineering tasks. In addition to mere code generation, DeepSWE exhibits sophisticated emergent behaviors; when confronted with bugs or feature requests, it engages in critical reasoning regarding edge cases, searches for existing tests in the codebase, proposes patches, creates additional tests to avert regressions, and adapts its cognitive strategies based on the specific challenges presented. This remarkable adaptability and efficiency position DeepSWE as a formidable asset in the software development landscape, empowering developers to tackle complex projects with greater ease and confidence. Its ability to learn from each interaction further enhances its performance, ensuring continuous improvement over time.