Meta presented Muse Glimmer, an AI model designed to run agent tasks directly on computers as Zuckerberg reinforces his commitment to open models.
Meta Platforms puts the open source AI at the heart of its strategy, the company unveiled Muse Glimmer, a new artificial intelligence model smaller than some of the market-leading systems and ready to run agent tasks on a Mac or PC equipped with a single graphics card.
At the same time, Mark Zuckerberg he advocated for greater openness in the development of artificial intelligence in the United States. The CEO of Meta believes that reducing barriers to open models will be important to compete with Chinese developers who are gaining ground in this sector.
The strategy represents a new move by Meta to regain prominence in the global AI race after the reception of Llama 4, and while US companies such as OpenAI, Anthropic y Google they are keeping the weights of their main models closed.
Muse Glimmer brings artificial intelligence to the device
Unlike larger models that rely on a considerable computing infrastructure, Muse Glimmer is designed to run directly on personal computers with a single GPU.
This approach can reduce dependence on external servers for certain tasks and opens up possibilities for developing AI agents that run locally on users’ computers.
Furthermore, the smaller size of the model reflects a growing trend in the sector to develop systems capable of performing advanced tasks without always requiring enormous resources data centers.
The ability to download, modify, and run models also offers greater flexibility to developers and companies that need to adapt artificial intelligence to specific applications.
Zuckerberg is pushing for more open AI models
Furthermore, Zuckerberg accompanied the launch with a 14-page essay titled The future is for everyone where he questioned why the development of AI is concentrated in just a few companies.
The idea that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic.
Mark Zuckerberg CEO de Meta.
The CEO also announced that Meta it is preparing larger models, thus making it clear that Muse Glimmer represents part of a broader strategy around open-source AI.
Meta also plans to publish the weights of Muse Spark 1.2, described as its most advanced model and developed by the superintelligence team created by the company.
China increases pressure on open-source models
Meanwhile, competition is intensifying from China, with companies like Moonshot AI, Alibaba and DeepSeek they have developed open models capable of competing in performance with US systems.
Among them appear Kimi K3 from Moonshot, Qwen3.8-Max from Alibaba and V4-Flash from DeepSeek, its progress is reinforcing China’s role within the open models ecosystem.
In contrast, the main systems developed by OpenAI, Anthropic and Google they keep their weights closed. This limits the degree of access and modification available to external developers compared to open alternatives.
For Meta, that difference represents an opportunity, the company has advocated a more open strategy for years, and its Llama models helped spread this approach among developers before Llama 4 forced a review of some of its plans.
Costs and cybersecurity are driving the debate about open AI
Furthermore, the interest in open models is related to the increasing cost of developing and using them artificial intelligence systems with readily available components can be more economical and allow organizations to customize their operation.
Cybersecurity is also part of the debate, the restrictions incorporated into certain closed models can limit some defensive uses, while open models offer greater scope for modification.
However, this freedom also increases the responsibility of those who implement these systems, access to the data allows for studying, adapting, and deploying a model with greater autonomy, but it requires adequate controls to manage its potential uses.
Meta will allocate up to $145 billion to AI infrastructure
Open-source software does not eliminate the need to build a huge physical infrastructure, Meta plans to invest up to $145 billion this year for AI infrastructure, according to the information provided.
Zuckerberg also announced a fund $1 billion intended to address concerns of communities affected by the construction of data centers.
The growth of these facilities faces resistance in various parts of the United States due to their impact on the land and local resources. For large technology companies, securing electrical capacity, land, and permits is becoming as important as developing better models.
Thus, Meta’s strategy combines two fronts, one is expanding access to artificial intelligence software and the other is increasing the infrastructure needed to develop increasingly capable models. Muse Glimmer represents the most accessible part of this approach, bringing agent tasks to personal computers while the company prepares larger-scale systems.
Source: Reuters
Photo: Shutterstock
