Meta's Muse Spark Open Targets Agentic Coding

Meta AI has released Muse Spark Open, an open-weight model built for agentic coding and released under the company's Llama license. According to Meta's research blog, the model is positioned as the most capable US open-weight model to date, with an emphasis on reasoning and software development tasks rather than general chat.

The headline feature is a 1-million-token context window, which gives the model room to hold entire codebases, long documentation sets, or multi-step agent traces in a single session. That scale matters for agentic workflows, where a model needs to track state across many tool calls and file edits without losing the thread.

Why it matters

Open-weight releases from a major US lab remain comparatively rare, and the agentic-coding framing puts Muse Spark Open squarely in competition with the coding-focused models that developers increasingly reach for. Key points from the release:

• Open weights under the Llama license
• A 1M-token context window aimed at long-horizon tasks
• Modalities spanning code, text, and reasoning

The permissive-but-conditional Llama license means teams can self-host and fine-tune, though the usual Llama terms apply. For developers building agents that plan, write, and revise code over long sessions, a large-context open model is a meaningful addition to the toolkit.

Meta has not published full parameter counts or benchmark figures in the record available here, so independent evaluation will be the real test of the "most capable" claim. Still, the combination of open weights and a very large context window gives the community something concrete to build on.