Meta Releases Muse Glimmer With Apache 2.0 Licence, Muse Spark 1.2 Open Weights Coming

Meta is releasing the weights for Muse Glimmer, a 30-billion-parameter dense AI model designed to run locally on a laptop or a single consumer GPU, as the company moves back toward open-weight artificial intelligence models. Meta is also preparing to release an open-weight version of Muse Spark 1.2, one of its leading foundation models, in the coming weeks.

Muse Glimmer is being released under the Apache 2.0 licence, with its weights available through Hugging Face. Meta says the model is intended for local agents, function calling, coding and LLM-as-a-judge workloads without requiring a cloud connection. The company is positioning the model as a way to bring agentic AI capabilities to consumer hardware rather than keeping them exclusively on cloud infrastructure. Mark Zuckerberg Lays Off 8,000, Spends USD 130 Billion on AI and Then Says AI Has Created a Lot of Jobs: Report.

Meta Releases Muse Glimmer, Muse Spark 1.2 Open Weights Coming

Muse Glimmer Targets Local AI

Muse Glimmer is a 30-billion-parameter dense model focused on agentic workloads. Meta says it can run on 24GB of VRAM without losing agentic reliability, potentially putting the model within reach of systems equipped with a high-end consumer GPU.

Alexandr Wang posted on X: "Big announcement today: we will be releasing an open weight version of muse spark 1.2 soon. we also are releasing muse glimmer, a 30B agentic model with open weights under apache 2.0. muse glimmer can run on 24GB of VRAM without losing agentic reliability." Meta Accelerates AI App Development, More Products Coming.

"Just like much larger models, muse glimmer can operate as a fully capable agent via planning, tool calls, checking its own results, and failure recovery. The model is designed to handle more than conventional text-generation tasks. Its intended applications include local AI agents, function calling, coding and LLM-as-a-judge workloads," he added.

Focus on Agentic Capabilities

Meta's pitch for Muse Glimmer centres on its ability to perform multi-step tasks rather than simply generate responses to individual prompts.

According to Wang, the model can plan tasks, make tool calls, check its own results and recover from failures. Those capabilities are increasingly central to the development of AI agents that can complete tasks with less human intervention.

The ability to run such workloads locally could also reduce reliance on cloud-based AI services for certain applications, although the practical experience will depend on hardware, model configuration and workload.

Muse Spark 1.2 Weights Coming Soon

Alongside Muse Glimmer, Meta is preparing to open the weights for Muse Spark 1.2.

The planned release represents a significant shift from Meta's recent strategy with the Muse family. Muse Spark was introduced in April as the first model from Meta Superintelligence Labs and as a natively multimodal reasoning model with support for tool use and multi-agent orchestration.

Meta's current developer materials describe Muse Spark as supporting end-to-end agentic workflows, advanced coding capabilities and native multimodal perception. The model is available through Meta's Model API for developers in the US.

Meta Returns to Open-Weight Strategy

The Muse Glimmer announcement marks a renewed emphasis on open weights from Meta as the company competes with other major AI developers.

Open-weight models allow developers and researchers to obtain model weights and run them on their own infrastructure, giving them greater control over deployment than a conventional cloud-only API.

Muse Glimmer's Apache 2.0 licensing is particularly notable because the permissive licence can allow developers to use, modify and redistribute software and model components subject to its terms.

Meta has a long history of releasing AI models and research openly, although its newer Muse Spark family initially followed a more proprietary approach. Meta's AI platform continues to describe its model strategy as focused on making AI capabilities available to developers and building agentic systems.

Why Local AI Matters

Running AI models locally can offer advantages including greater control over data, reduced dependence on internet connectivity and potentially lower recurring inference costs for some workloads.

The challenge is that larger models can require substantial memory and computing resources. Meta's claim that Muse Glimmer can operate with 24GB of VRAM is therefore significant for developers and enthusiasts who want to experiment with agentic AI without operating a large cloud infrastructure setup.

However, the model's real-world performance will depend on the hardware and optimisation techniques used by developers. A 30-billion-parameter model remains considerably more demanding than smaller local models.

What Comes Next for Meta's Muse Line

The planned Muse Spark 1.2 open-weight release could broaden access to Meta's more capable foundation models, while Muse Glimmer provides a smaller option aimed specifically at local agentic workloads.

Meta's recent Muse releases have expanded beyond language models into image and video generation. The company introduced Muse Image and Muse Video in July as media-generation models developed by Meta Superintelligence Labs.

The two upcoming open-weight moves indicate that Meta is seeking to combine its proprietary model development with a broader developer ecosystem around models that can be downloaded and deployed independently.

For developers, the key question will be how Muse Glimmer performs against other local models and how the forthcoming Muse Spark 1.2 weights compare with leading open-weight foundation models once they become available.

(The above story first appeared on LatestLY on Aug 10, 2026 05:25 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).