Groq announced that Meta’s Muse Spark Open is available in limited preview on GroqCloud on launch day. The open-weight model is designed for agentic coding, and Groq says it serves more than six million developers, Fortune 500 enterprises and thousands of AI-native companies.

Muse Spark Open has 816 billion total parameters, with 42 billion active per token, and uses a Mixture of Experts design with 256 experts. GroqCloud supports a 131,072-token context window, text and image input with text output, and speeds of 350+ tokens per second per user. The model was trained on developer workflows including multi-file refactors, debugging sessions and long-running tasks.

Meta designed Muse Spark Open to plan, act, observe and correct course over extended tasks. It can spawn and coordinate subagents, browse the web, and use computers across desktop, mobile and browser environments. Before each action, it writes a short statement of intent. Meta reports higher first-attempt accuracy and reliable tool calling.

Groq says agentic coding tasks involve chains of model calls, with each step waiting for the previous one, so inference speed affects how long developers wait for results. Groq said its platform processes trillions of tokens weekly in data centers across North America, Europe, the Middle East and APAC. Groq executive Scott Albin said agents make dozens of model calls on a single task, and running Muse Spark Open on GroqCloud keeps those calls fast.

Groq announced that Meta’s Muse Spark Open is available in limited preview on GroqCloud on launch day. The open-weight model is designed for agentic coding, and Groq says it serves more than six million developers, Fortune 500 enterprises and thousands of AI-native companies.

Muse Spark Open has 816 billion total parameters, with 42 billion active per token, and uses a Mixture of Experts design with 256 experts. GroqCloud supports a 131,072-token context window, text and image input with text output, and speeds of 350+ tokens per second per user. The model was trained on developer workflows including multi-file refactors, debugging sessions and long-running tasks.

Meta designed Muse Spark Open to plan, act, observe and correct course over extended tasks. It can spawn and coordinate subagents, browse the web, and use computers across desktop, mobile and browser environments. Before each action, it writes a short statement of intent. Meta reports higher first-attempt accuracy and reliable tool calling.

Groq says agentic coding tasks involve chains of model calls, with each step waiting for the previous one, so inference speed affects how long developers wait for results. Groq said its platform processes trillions of tokens weekly in data centers across North America, Europe, the Middle East and APAC. Groq executive Scott Albin said agents make dozens of model calls on a single task, and running Muse Spark Open on GroqCloud keeps those calls fast.