Meta Releases Open Source, Open Weight Version of its Most Powerful AI Model

Meta today announced Muse Glimmer, an open source and open weight AI model that’s optimized for local agent workflows and is based on its most powerful closed AI model, called Muse Spark. Its release was accompanied by a 6,500 word blog post allegedly penned by Meta co-founder and CEO Mark Zuckerberg in which he argues that superintelligence requires a “philosophy based on individual empowerment.” Given his lack of intellectual depth and self-awareness, one can only assume that Meta’s AI strategy isn’t working, so the company is pivoting to remain relevant in this space.

“Today, we’re introducing Muse Glimmer, the next model from Meta Superintelligence Labs, and open sourcing the model weights under a permissive Apache 2.0 license,” the Meta announcement explains. “Muse Glimmer is a 30-billion-parameter model optimized for always-on local agent workflows. It’s small enough to run on a Mac or PC with a single consumer GPU, enabling use cases that range from local agents and function calling, to local coding, and LLM-as-a-judge evaluation. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category.”

Meta says that Muse Glimmer is trained for end-to-end agentic task completion, reliable tool use, multi-step reasoning, failure recovery, multimodal input and reasoning, OpenClaw and other scaffold compatibility, automatic speed and quality selection, and multilanguage support. But this is a local AI model, not a cloud-based frontier model, so it compares favorably against similar but lesser-known local AI models while being optimized to work well on consumer computer hardware.

The Zuckerberg manifesto is … interesting. And even more so the more you know about this person. His company’s products are so privacy invasive that they inspired Apple to create App Tracking Transparency to block all the tracking, and so his call to build a superintelligence that “empowers everyone” and “improves all our lives” feels more than a bit hypocritical.

“Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it,” he writes. “This has the potential to begin a new era of personal empowerment where individuals can use this powerful new capability to reach their full potential, pursue their interests, and improve their lives and the world more than ever before.”

It’s unclear what the business model is here, but Meta will spend $145 billion this year building datacenters and plans to spend up to $600 billion on this effort by 2028. This spending includes paying off locals to overcome their opposition to datacenters through a $1 billion Future Is For Everyone Fund. Yes, really. Zuckerberg also has policy advice for the U.S. government and says the country is lagging behind China because of its massive investments in nuclear energy capacity to power its AI infrastructure.

You can get started with Muse Glimmer today via Hugging Face, with Ollama, LM Studio, and Unsloth availability coming soon. Meta has also published documentation for the model in its AI Developer Center.