Meta’s Muse Glimmer marks a new open-model push — and a bigger compute bet
Meta has released Muse Glimmer, a 30B open agent model from its new Superintelligence Labs, while Mark Zuckerberg argues for broader model access, distillation, and a possible auction-based compute business.
Meta reopens its model-weight strategy with Muse Glimmer
Meta has released Muse Glimmer, a 30-billion-parameter agent model with weights available under an Apache 2.0 license on Hugging Face. The release is Meta’s first open model since Llama 4 in spring 2025 and the first open model from Meta Superintelligence Labs. According to the Wall Street Journal, an open-weight version of Muse Spark 1.2, described as Meta’s strongest model right now, should follow in the coming weeks.
Glimmer is built for local agents, not just cloud demos
Muse Glimmer is designed for AI agents that run locally around the clock on a Mac or PC with a single consumer GPU. At full precision, the model would need more than 55 GB of memory, but Meta compresses the weights to about 4 bits, bringing memory needs under 20 GB. Meta says a small helper model can speed up text output by up to 3.1x, while the product pitch emphasizes keeping calendar, file, and message handling on-device rather than sending data to the cloud.
Benchmarks show Meta is back in the mix, not clearly ahead
Meta compares Glimmer with Google’s Gemma4-31B and Alibaba’s Qwen3.6-27B, described as leading open models in the same size class. Glimmer wins most of Meta’s reported benchmarks, especially on agent tasks such as tool use, web search, and long-context work. The article cautions that Meta gathered most of the comparison data itself and says the test setup was not tuned for rival models, so the figures should be read carefully.
Zuckerberg turns distillation into a policy argument
Alongside the release, Mark Zuckerberg published an essay titled “The Future is for Everyone,” arguing that superintelligence should be widely distributed rather than controlled by a small number of labs. He also defended distilling other companies’ models, writing that the principle worth protecting is that “you can learn from anything you can observe.” That position directly contrasts with concerns from OpenAI and Anthropic, which have accused Chinese labs of using their models as teachers without permission and raised alarms about frontier-level open models.
The business question: can Meta monetize free models through compute?
The article frames Meta’s open-model strategy against major planned infrastructure spending, including up to $145 billion in investments this year and $600 billion through 2028, per the Wall Street Journal. Open models like Glimmer do not generate licensing revenue by design, and Meta does not have an API business at a comparable scale to leading rivals cited in the article. Zuckerberg’s essay points to a possible “dynamic auction mechanism” where free versions reach billions of people, while users who want more compute pay based on demand and scarce data-center capacity.