Mark Zuckerberg Open Letter on Meta’s Personal Superintelligence

Meta has published a new open letter from Mark Zuckerberg laying out a broad vision for “personal” superintelligence—an AI future meant to be widely accessible rather than locked behind a small group of companies—and paired it with two new open-source releases from its Muse model family.

Announcement: “superintelligence” as a personal tool

Zuckerberg argues that the most advanced AI capabilities shouldn’t be concentrated among a few organizations or institutions. In his view, superintelligence should function as a personal tool that helps individuals create, learn, invent, and contribute to scientific progress.

He also frames the moment as historically significant and says the opportunities and challenges will be larger than anything seen in a person’s lifetime. Zuckerberg calls for a clear philosophy on how superintelligence is developed and distributed, emphasizing that security shouldn’t rely only on control by a small number of actors, but also on how power is shared.

In the same letter, he questions the idea that AI will inevitably lead to large-scale job loss and human irrelevance, and he positions superintelligence primarily as a way to boost individual capabilities and increase invention—rather than replacing people through automation.

Practical impact: open access, local-first capabilities, and “agents”

Meta’s target is to push advanced capabilities into the hands of individual users through open approaches. The “personal superintelligence” concept described in the letter is built around tools and agents that understand what users want to achieve, then help them across many kinds of tasks.

Meta lists several components such a setup could include: a personal agent that understands user needs and goals; tools to turn ideas into real outputs (from content creation to project execution); tools for starting new businesses; a personal tutor or coach for learning; and tools to support scientific research, including open biological models that could help with drug design.

Meta also says these capabilities should be available for free or at prices people can afford, to avoid superintelligence becoming a privilege limited to those who can pay.

Platform scope: what Meta says about running Muse locally

Alongside the letter, Meta announced Muse Glimmer with a strong local-execution focus. The company says Glimmer is designed to run on a laptop or a single consumer GPU and can operate fully on the device without requiring data to be sent to a cloud service.

Meta claims broad support across the local AI stack, naming hardware ecosystems such as NVIDIA, AMD, Intel, Arm, and Dell. For local inference tools and runtimes, it mentions Ollama, LM Studio, llama.cpp, and MLX. For inference and serving platforms, Meta cites vLLM, SGLang, Together AI, Fireworks AI, and OpenRouter.

For recommended hardware, Meta mentions a Mac mini or MacBook with an M4 or M5 chip and at least 32 GB of memory, or an NVIDIA RTX 5090 GPU / AMD Radeon AI PRO.

What changed: new Muse open-source model weights

Muse Glimmer weights released: a 30B-parameter model positioned for local execution on a laptop or a single consumer GPU.
Muse Spark 1.2 weights scheduled: Meta says weights will be available in the coming weeks for a new version it describes as among the most advanced foundation models in its category.
Local-first positioning for Glimmer: Meta states the model can run entirely on-device and is intended for complex, multi-step tasks without step-by-step guidance.

What comes next

Meta’s next step is the rollout of Muse Spark 1.2 weights in the coming weeks. In parallel, the company’s letter emphasizes that the broader goal is to make advanced AI capabilities usable outside proprietary infrastructure—supporting developers and individuals who want to build with these tools.