Meta Unveils Open-Weight AI Model to Challenge Chinese Rivals
Meta on Thursday launched Muse Spark Open, an open-weight AI model it claims rivals and in some cases outperforms products from leading Chinese developers. The model, described as “the most capable US open-weight model to date” by Meta’s AI chief Alexandr Wang, allows developers to download and modify its parameters, contrasting with closed models like OpenAI’s ChatGPT or Anthropic’s Claude.
Collaboration with Apple Enables Desktop Supercomputing
The launch coincides with a partnership between Meta and Apple, enabling Muse Spark Open to run on Apple’s M5 Ultra chips. Meta AI chief Alexandr Wang highlighted that Apple’s unified memory architecture allows the M5 Ultra to handle the model’s demanding computational needs, with two machines offering up to 512 GB of memory per unit.
Technical Benchmarks and Competitive Claims
Meta positioned Muse Spark Open for coding, agentic work, and multi-step reasoning, emphasizing “strong safeguards” against risks. The model was optimized using MLX, a framework developed by Meta, and is available via Hugging Face. While the standard Muse Wonder model is accessible through an API, its more powerful variant, Muse Wonder Full, is reserved for paid subscribers. Benchmarks from Artificial Analysis ranked Wonder Full at 56 on its Intelligence Index, tying with Anthropic’s Claude models.
Meta has released Muse Wonder and Muse Wonder Full, achieving Artificial Analysis Intelligence Index v4.3 scores of 52 and 56 at max reasoning effort – each a material gain over Muse Spark 1.3 (max) at 48, with a release gap of only 6 weeksMuse Wonder is available for users on… pic.twitter.com/85mvsVA0ii
Context of Meta’s AI Strategy Shift
The release follows Meta’s recent focus on closed models, including a new AI system announced days prior that it claimed rivals ChatGPT and Claude. However, the company’s AI efforts initially prioritized open models, with Wang leading the charge at Meta Superintelligence Labs. The shift reflects broader industry competition, as Chinese developers like Alibaba’s Qwen and Moonshot AI’s Kimi K3 have dominated the open-weight market.
Apple’s Growing Role in AI Hardware
Apple’s collaboration with Meta shows its expanding influence in AI chip development, a sector traditionally dominated by Nvidia. The partnership could signal Apple’s growing strength in AI chips, as Apple’s M5 Ultra chips offer high memory bandwidth and performance for local AI workloads. This aligns with Apple’s recent moves, including a 2024 deal with OpenAI to integrate ChatGPT into Siri, though that collaboration faced performance issues.

Unanswered Questions in the AI Race
While Meta emphasizes Muse Spark Open’s capabilities, questions remain about its practical adoption. The long-term impact of Meta’s dual strategy—balancing open and closed models—on market dynamics and user trust remains unclear. As the AI field evolves, the interplay between open-source innovation and proprietary systems will likely define the next phase of competition.
More on Muse Spark Open
Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work. pic.twitter.com/NKHrybrGKQ
meta is just teasing us that they have prerelease M5 Ultra 512GB unified memory Mac Studios https://t.co/j7QI8fkiQX pic.twitter.com/089eWrL2mv
We're open-sourcing Muse Spark Open!This is an important release which is the most capable US open-weight model yet. 🇺🇸- 816B parameters (42B active)- 1M token context window- Built for coding and long-horizon agents- Runs locally on two Apple M5 Ultra Mac StudiosMore… pic.twitter.com/uDmhiWI1hN
Meta’s Muse Spark Open is an 816B open-weight model that beats US rivals and matches Chinese models on agentic tasks, but its license isn’t fully open. https://t.co/CM1lum1l9f
Muse Spark Open — weights releasing today — runs great on new Mac Studios thanks to work with @johnternus and Apple. pic.twitter.com/U2wJNrKBOg
love that Meta open-sourced Muse Spark Open!it's ~Muse Spark 1.2 level.my biggest wish: Meta follows up with Muse Spark 1.3 open weights.that would be the best OSS model in the world.
Meta is so back! They are finally releasing Meta Muse Spark open weight.Best US open weight model 816B total / 42B active parameters1M token context windowNatively multimodal across text, image, video, audioLove to see! https://t.co/zBoXFotUUu pic.twitter.com/dcKMngCUl9
To help make advanced models more accessible, Muse Spark Open is optimized to run on two Apple M5 Ultra Mac Studios with the latest MLX framework.It also runs on leading inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth with…
Keep reading