Muse Spark Open is Meta’s answer to Qwen, GLM, and Kimi in the open-weight race

What it does

Meta launched Muse Spark Open, a new large open-weight AI model with 816 billion parameters. It handles text, images, video, and audio inputs, supporting up to one million tokens of context. Meta specifically optimized it to run efficiently on Apple’s M5 Ultra chip, aiming to enable powerful local AI workflows. Performance benchmarks put it roughly on par with leading Chinese open-weight models like Qwen, GLM, and Kimi.

Why it matters

Muse Spark Open tightens the competition in open-weight AI models by offering massive capacity and multimodal support accessible without cloud lock-in. Local execution on M5 Ultra chips can lower latency, cut cloud costs, and improve privacy for businesses and developers working with large context windows. Matching top Chinese models benchmarks means Meta is seriously pushing its AI ecosystem against well-funded global rivals.

Who it is for

This model directly targets builders and companies aiming for high-end AI tasks with mixed media inputs while avoiding dependence on external cloud services. Apple users and enterprises with on-prem computing setups will see immediate advantages deploying Muse Spark Open. Those working in AI research or building multi-turn conversational agents with rich context should also find its scale and modality support valuable.

The catch

Muse Spark Open remains a heavyweight at 816 billion parameters, requiring advanced hardware like Apple’s M5 Ultra to run locally, which limits accessibility for smaller operators. Although it competes well on benchmarks, practical deployment costs and engineering complexity may slow adoption. There is also limited information on fine-tuning capabilities, inference speed, and ecosystem integrations beyond the Apple hardware niche.

What to watch next

How widely Meta opens access to Muse Spark Open will determine its real-world impact. Tracking adoption among Apple-based developers and enterprises will show if local high-capacity models can chip away at cloud AI dominance. Watch how competitors respond, either by boosting multi-modal scale or improving local hardware compatibility. Also, observe how open-weight options challenge the closed proprietary models that currently dominate the market.

AI Quick Briefs Editorial Desk