Muse Spark Open, an open-weight model for coding agents

Meta has released Muse Spark Open, an 816-billion-parameter open-weight omnimodal mixture-of-experts model with 42.7 billion active parameters, a one-million-token context window, and support for text, image, video, and audio inputs with text output. Co-trained with Muse Code for coding-agent workflows, it can support tasks such as screenshot-based UI debugging, converting bug videos into reproducible tests, and voice-driven programming, while also integrating with third-party agents including OpenCode, Pi, and Deep Agents Code. Meta-reported benchmarks position it competitively against frontier models in routine agentic coding and knowledge-work tasks, though the comparison uses each vendor’s own agent framework and suggests it may require more guidance on large multi-file changes than some closed competitors. Its architecture uses compressed MoE layers to reduce model size, inter-GPU communication, and memory demands, alongside a hybrid attention system of local and global layers that reportedly reduces KV-cache memory by about 3.7 times at 128K context. Baseten is offering the model through OpenAI-compatible Model APIs and Dedicated Inference using an NVFP4 checkpoint that reduces weight storage from roughly 1.6 TB to 540 GB, allowing the model to run on a single four-GPU GB300 node and reserve more memory for concurrent long-context agent sessions.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.