Meta opened the weights of its 30-billion-parameter Muse Glimmer artificial intelligence model on August 10, while CEO Mark Zuckerberg said the company would open the weights for its latest AI model, Muse Spark 1.2, meaning it can be downloaded and used by the public.
Meta said it would open source its most powerful AI models and launch new ones designed for consumer devices, as it looks to rival leading labs like OpenAI and Anthropic.
What Muse Glimmer is and what it can do
Muse Glimmer is an open-weight model designed for agentic tasks that can run on a Mac or PC with a single graphics card, aiming to tap demand for AI systems that run directly on people’s devices.
It gives developers and enterprises a 30-billion-parameter model optimized for local agentic workflows, published under a permissive Apache 2.0 license, with weights available on Hugging Face. Meta says the model supports text and image inputs, covers over 100 languages, and is compressed to under 20 GB for efficient operation on standard hardware, targeting developers building AI agents and coding tools.
Meta said it is also working with AMD, Arm, Dell, Intel and NVIDIA to further optimize the model’s performance across devices.
Muse Glimmer is an open-weight model designed to power AI agents locally on consumer hardware. The 30-billion-parameter model is essentially an open version of Meta’s most powerful closed model, Muse Spark. Glimmer’s weights are available under the permissive Apache 2.0 license, so developers can download and modify them as necessary.
That distillation approach mirrors a wider industry pattern in which labs compress frontier-class reasoning into models developers can run locally without a cloud subscription, an area also explored in Google DeepMind’s on-device SL2T model for Pixel 11.
The Muse Spark 1.2 open-weight pledge
Meta said it would open source its most powerful AI models and launch new ones designed for consumer devices. In an Instagram video posted Monday, CEO Mark Zuckerberg said that the company would open the weights for its latest AI model Muse Spark 1.2, meaning it can be downloaded and used by the public.
Muse Code is Meta’s terminal coding agent, launched in beta on August 5, 2026 and powered by the new Muse Spark 1.2 model. It features parallel sub-agents, worktree isolation, and a crash-safe event log. Muse Spark 1.2 is available through Muse Code and the Meta Model API with expanded global access. Meta also linked an OpenRouter page for the model at launch.
Muse Spark 1.2 was closed at launch, but on August 10, 2026, alongside the Muse Glimmer open-weight release, both Zuckerberg and Alexandr Wang said on X that open weights for Muse Spark 1.2 are coming “soon.” No date has been confirmed, and Muse Code’s own open-source status remains unresolved separately.
Muse Spark 1.2 scores 54 on the Artificial Analysis Intelligence Index, level with Grok 4.5. Meta has reported scores of 51 for version 1.1 and 43 for the original model, though those figures have not been independently confirmed.
Zuckerberg’s policy argument
In a 6,500-word essay on AI published Monday, Zuckerberg positioned Meta’s move as a challenge to Chinese open-source technology and urged Washington to support American efforts.
Zuckerberg wrote that the goal should be for American open-source models to be the best globally, and called for removing hurdles that make it harder for American open-source models to compete. He added that he does not believe restricting access to foreign open-source models is an effective solution. Zuckerberg said the U.S. will need to rethink its policies in several areas, including distillation and data use in training, distillation broadly referring to the use of output from an advanced AI model to train another model, a practice that has attracted controversy, with some U.S. lawmakers viewing it as effectively theft of intellectual property.
The announcement marks the latest show of support for open-weight AI, which is gaining traction as businesses grow wary of ballooning AI bills and worry about recent cybersecurity incidents involving models from Anthropic, OpenAI and Meta.
The release reinforces Meta’s long-running argument that powerful AI models should be made accessible to a broad developer base rather than being controlled by a small number of technology companies.
Market and developer implications
Shares of Meta were up 2.1% in premarket trade on Monday. The model has already been integrated into Meta’s AI products, including its Meta AI application and website.
Meta has also been expanding Muse Spark across its broader ecosystem, including WhatsApp, Instagram, Facebook, Messenger and AI-powered glasses.
For developers weighing infrastructure costs, running a frontier-distilled model locally avoids per-token API fees entirely, a calculation that becomes especially relevant as AI hardware costs continue to climb across the supply chain.
Muse Code is Meta’s answer to Claude Code, OpenAI’s Codex, and Google’s Antigravity CLI, and is the first coding-specific product to ship from Meta Superintelligence Labs.
Five days after Muse Spark 1.2 launched, Meta said it will open the model’s weights, an announcement that, if it lands, would make the model the strongest current U.S. open-weight rival to Chinese models. Muse Spark launched in April 2026 as the first model from the reorganized Meta Superintelligence Labs. The lab is headed by Alexandr Wang, Meta’s first-ever chief AI officer.
If the weights ship as pledged, Meta will have made its entire current frontier model line available to any developer with a capable GPU.
