Meta Muse Spark Open launches on Together AI with 1M token context

Together AI is launching Meta's open-weight Muse Spark Open as a day-one partner, positioned for long-running agentic coding workflows. Its 1M token context window can hold entire projects with complex dependencies, legacy code, and thousands of files; it supports long-horizon autonomous planning, execution, observation, and self-correction, and can coordinate sub-agents to split tasks. It also has computer-use capabilities across desktop, mobile, and browser plus multimodal browsing, accepting text, image, audio, and audio-video input, and states intent before acting for auditability. The model was trained on real developer work, including multi-file refactors, debugging sessions, and long tasks. Pricing is $1.25 per million input tokens, $4.25 per million output tokens, and $0.15 per million cached input tokens, callable serverlessly via the Together API under the model ID meta-models/Muse-Spark-Open-816B-A42B. Early-access partners Factory and Ollama both offered the model to their users on launch day. For teams running agentic coding workflows on Together AI, this means fitting an entire codebase into a single context, reducing chunking, and cutting repeated-prompt costs via cached input; Ollama says the model performs well on multi-step and parallel tool calls, long context, and structured output.