Meta Launches Its Most Powerful US Open-Weight AI Model, Muse Spark Open

Meta has introduced Muse Spark Open, an open-weight AI model designed for coding, complex reasoning and agentic tasks, as it aims to strengthen its position in global AI competition.

Meta Unveils Muse Spark Open AI Model

Meta has announced the launch of Muse Spark Open, its latest open-weight artificial intelligence model, as the company seeks to compete more strongly in the rapidly evolving AI market.

Alexandr Wang, Meta’s Chief of AI, announced the release on X, describing the model as the most capable open-weight AI model developed in the United States to date.

According to Meta, Muse Spark Open is designed to compete with leading open-weight models from international developers, including companies based in China. The launch reflects Meta’s continued efforts to expand its AI capabilities and provide developers with more flexible tools for building applications.

What Are Open-Weight AI Models?

Open-weight AI models allow developers to access and download a model’s trained parameters, commonly referred to as weights. Depending on the model’s licence and release conditions, developers may be able to customise, modify and run the model for different applications.

This approach differs from many closed AI systems, where users generally interact with the technology through a hosted service or API without receiving direct access to the model weights.

Popular closed AI offerings include Anthropic’s Claude and OpenAI’s ChatGPT services. Open-weight models, meanwhile, can give developers greater control over deployment, customisation and integration into their own systems.

However, open weights do not automatically mean that every part of a model is open source. Access to training data, source code and other development details can vary between releases.

Meta Targets Competition From Chinese AI Developers

Chinese technology companies have gained significant attention in the open-weight AI market through models designed for reasoning, programming and general-purpose applications.

Examples include Alibaba’s Qwen family, Moonshot AI’s Kimi models and Z.ai’s GLM series. These offerings have contributed to growing competition in a field where developers increasingly have alternatives to proprietary AI platforms.

Meta says Muse Spark Open can match or outperform comparable models from leading Chinese developers. The claim positions the new release as an attempt to strengthen the United States’ presence in the open-weight AI ecosystem.

Actual performance comparisons, however, depend on the benchmarks used, the tasks being tested and the computing resources available. Independent evaluations will be important in determining how the model compares with competing systems across different workloads.

Muse Spark Open Focuses on Coding and Agentic Tasks

Meta says Muse Spark Open has been developed with a particular emphasis on coding, software engineering and agentic work.

Designed for Complex Software Development

The model is intended to support demanding programming tasks, including complex software engineering projects that may require several stages of planning, execution and review.

Such capabilities can help developers use AI tools for more than generating short code snippets. Models designed for longer tasks may assist with understanding requirements, solving technical problems and managing interconnected steps in software development.

Multi-Step Reasoning and Long-Horizon Tasks

Muse Spark Open is also designed to handle long-horizon tasks, which involve working towards a goal through multiple stages rather than responding to a single, straightforward prompt.

Multi-step reasoning is particularly relevant to AI agents that need to analyse information, make decisions and complete sequences of actions. These capabilities are becoming increasingly important as technology companies develop AI systems for more complex professional and technical workflows.

Meta Highlights Safety Measures

Alongside its performance claims, Meta says Muse Spark Open incorporates strong safeguards.

Safety measures are important for AI models that may be used for coding, automated workflows and other complex activities. Such protections can help reduce the risk of harmful outputs or misuse, although their effectiveness depends on the model's design, deployment environment and the safeguards applied by developers.

For organisations adopting open-weight AI, security and governance considerations remain important because the ability to run and modify models can create different risks compared with using a centrally managed AI service.

Muse Spark Open Is Optimised for Apple M5 Ultra Chips

Meta has also highlighted optimisation for Apple’s M5 Ultra chips. Hardware optimisation can influence how efficiently an AI model runs, including its speed and computing requirements.

Compatibility with powerful hardware could be relevant to developers and organisations seeking to run AI workloads locally or in dedicated computing environments.

However, the practical benefits will depend on the model’s hardware requirements, software support and performance under real-world workloads.

New Release Follows Meta’s Closed AI Model

The launch of Muse Spark Open comes shortly after Meta introduced another AI model with a closed-weight approach. The company said that model could compete with many leading frontier AI systems.

The two releases illustrate Meta’s efforts to pursue different approaches to AI development: proprietary systems and models whose weights can be accessed by developers.

Offering models under different access arrangements can help a company serve a broader range of users, from consumers and businesses to researchers and software developers.

Conclusion

Meta’s Muse Spark Open marks another development in the competition to build advanced AI models that developers can access and customise. With a focus on coding, agentic workflows and complex reasoning, the model aims to challenge established open-weight offerings. Its real-world impact will depend on independent performance assessments, safety evaluations and how effectively developers adopt it.