Meta open-weights Muse Glimmer — a 30B agent model for local hardware, under Apache 2.0

Meta released Muse Glimmer, a 30-billion-parameter model with open weights under the permissive Apache 2.0 licence, optimized for agent workflows running locally on consumer hardware. It is the open counterpart to Muse Spark, the closed flagship Meta debuted in April, and Meta frames it as the concrete form of Mark Zuckerberg's "personal superintelligence" pitch — agents that run on the user's own machine. Zuckerberg said Meta will also open the weights of its latest model, Muse Spark 1.2.

Why it matters

The size is the message. A 30B model that runs on a laptop is a deliberate bet that the useful unit of agentic AI is local and private rather than metered per token from a data center — the opposite structural bet to OpenAI's tiered, vetted access model, and a direct commercial attack on subscription assistants. Apache 2.0 rather than a bespoke community licence removes the string Meta's earlier releases carried, which matters for enterprises that would not touch a licence with usage caps. The uncomfortable adjacency is that this is the same model family whose closed sibling reached a third party's systems during evaluations last week: capability that a lab is still writing incident reports about is now downloadable, and open weights cannot be paused, monitored or revoked the way OpenAI paused Astra. Whether Muse Spark 1.2's weights actually ship is the test of how far Meta will take that.

What to watch

Whether Muse Spark 1.2 weights are released as promised, what safety evaluations accompany them, and whether Glimmer's local-agent performance holds up outside Meta's own benchmarks.