Meta Platforms has introduced Muse Glimmer, a new artificial intelligence model that users can download and run on personal computers, bringing more advanced AI capabilities within reach of people who do not want to rely entirely on cloud-based services.
The model, unveiled on Monday, is a streamlined version of Meta’s Muse Spark 1.2 and has been developed to require less computing power. At 30 billion parameters, Meta says Muse Glimmer can run on a single graphics card, making it possible for users with sufficiently powerful personal computers to operate the AI locally.
For someone using AI to organise files, manage a schedule or handle other routine digital tasks, that could mean having the technology work directly on their computer rather than sending every request to a remote server. Brandspur Brand News reports that the launch is also significant because Meta is making the model available without a download fee, although users remain responsible for the cost of the hardware and computing power needed to run it.
Meta said Muse Glimmer is primarily built for agent-like tasks, allowing it to carry out activities that require several steps rather than simply responding to individual questions. Its weights, which contain the learned values that guide how the model operates, will be made available through Hugging Face for users who want to download and customise the system.
The decision reflects Meta chief executive Mark Zuckerberg’s long-standing support for making AI models more openly available. In an accompanying essay, Zuckerberg argued that advanced AI should not be concentrated within a handful of companies or institutions, but should instead be made available to individuals.
That position puts Meta on one side of an increasingly heated debate over open-weight AI. Companies such as OpenAI and Anthropic have expressed concerns about the safety risks associated with releasing powerful models that people can freely download and modify. Meta, meanwhile, maintains that wider access can help individuals benefit directly from advances in artificial intelligence.
The competition is no longer limited to American technology companies. Google has its Gemma range of open-weight models, while Chinese firms, including DeepSeek, Moonshot and Alibaba, have released models that have increased pressure on US companies developing proprietary AI systems.
The growing influence of Chinese open models has also attracted attention in Washington. US policymakers have considered whether restrictions should be placed on some foreign open-weight systems, but Zuckerberg argued that limiting access to Chinese models would not be an effective answer.
Instead, he called for American companies to develop open models capable of competing globally and urged greater cooperation between the technology industry and government.
The stakes are substantial. Meta, Amazon, Alphabet and Microsoft have together committed more than $2 trillion to expanding their AI infrastructure and capabilities, as technology companies race to secure a leading position in the industry. The scale of the spending, however, has raised concerns among investors about whether the enormous investments will eventually deliver adequate returns.
Meta itself has been moving towards greater commercialisation of its AI products. It started charging developers for access to its Muse Spark 1.1 model last month and is also considering a cloud infrastructure business that could provide customers with access to computing power and AI models.
The company is simultaneously investing heavily in the physical infrastructure required to support its AI ambitions. It announced plans for a $1 billion fund targeting US communities where it operates data centres, as major technology companies face growing criticism over the land, electricity and water requirements of large data-centre developments.
Muse Glimmer therefore arrives at an important moment in the AI race. For developers, enthusiasts and everyday computer users with capable machines, the attraction is straightforward: a powerful AI model that can be downloaded, customised and used locally, without a fee for the model itself.
