Meta released Muse Spark Open on October 15, 2026, saying it is the most capable US open-weight model for coding and agentic tasks. It is the company's first open-weight model since its Llama series.
Meta reported Muse Spark Open scores well ahead of Qwen 3.8, GLM 5.3, and Kimi K3 on benchmarks like JobBench, MCP-Atlas, HLE-Diamond, and ZeroBench. That compares with Meta's paid Muse Wonder Full models, which it still falls short of.
Muse Spark Open is built for Apple's M5 Ultra and runs locally on two Mac Studios with 512 GB of unified memory each. Availability begins with weights on Hugging Face, initially for developers and researchers.
"Muse Spark Open is the most capable US open-weight model, built for coding and agentic tasks," said Meta. The model leads in Agentic, Knowledge & Reasoning, and Multimodal categories but falls behind some competitors on coding benchmarks.
The announcement follows Meta's release of the Llama series. Meta framed the significance of Muse Spark Open as its response to the growing competition in the open-weight space.
Meta did not say how the model will be monetized, and raised the open question of whether it can compete with its paid models. The company said more details are in the research blog.
Source: thedecoder
