
What happened
Meta released Muse Spark Open, an open-weight model with 816 billion parameters (42 billion active), up to one million tokens of context, and support for text, image, video, and audio. Meta says it scores well ahead of Qwen 3.8, GLM 5.3, and Kimi K3 in several benchmark categories, but still falls well short of Meta's paid Muse Wonder Full models.
Why it matters
Meta calls Muse Spark Open the most capable US open-weight model.
What to watch
Providers pulling in more than $20 million a year who want to offer Muse Spark Open as an API service need a separate agreement with Meta.
WHO IT HITSDevelopers and AI teams that self-host open-weight models gain a new option, with weights on Hugging Face under a free license that requires the label "Built with Muse Spark Open."
Summaries like this, in your inbox every morning.
Meta is returning to open-weight territory after its Llama series. The company is positioning Muse Spark Open as its answer to Chinese open-weight competitors such as Qwen, GLM, and Kimi, which have set the pace in that segment.
The model is optimized to run locally on Apple's M5 Ultra and on two Mac Studios with 512 GB of unified memory each, according to Meta. Weights are available on Hugging Face. Meta says Muse Spark Open leads in Agentic, Knowledge & Reasoning, and Multimodal categories but falls behind some competitors on coding benchmarks. The license is free to use with a "Built with Muse Spark Open" label, and providers earning more than $20 million a year that want to resell it as an API service need a separate agreement with Meta.
The benchmark comparison puts Muse Spark Open ahead of Qwen 3.8, GLM 5.3, and Kimi K3 in several categories, but Meta says the model still falls well short of its own paid Muse Wonder Full models. That gap suggests Meta is keeping its strongest capabilities behind a paid tier while offering the open-weight version as a competitive option for local use and coding tasks.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Nvidia launched Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model, plus Bolt 74B-A9B for l…
IBM outbound CTO Tyler Lynch said teams should start from customer outcomes, not roles
Japan Mobility Show Bizweek 2026 showcased Physical AI applications, including Toyota's basketball-playing rob…

Google pitched Gemini 4 Argon as one model for robot control, simulation and training-data review, saying it s…

Nvidia released Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters for enterprise age…

Perplexity is giving away SPACE, its "AI sandbox" for containing AI agents, as an open source project, with Co…
