{"slug": "meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop", "title": "Meta Releases Muse Glimmer, an Open-Weight AI Model That Runs on a Laptop", "summary": "Meta released Muse Glimmer on Monday, August 10, a 30-billion-parameter open-weight AI model that runs on a single consumer GPU, available for free download from Hugging Face under an Apache 2.0 license. CEO Mark Zuckerberg tied the release to a push for looser U.S. restrictions on training data, citing competition from open-source AI in China. The model is designed for agentic tasks like running local agents, coding assistance, and evaluating other models, and Meta claims it performs comparably to Google's Gemma 4-31B and Alibaba's Qwen 3.6-27B.", "body_md": "*Meta gave away a 30-billion-parameter AI model on Monday that runs on a single gaming laptop, and Zuckerberg used the moment to push Washington for looser rules on open-source AI.*\n\nMeta released Muse Glimmer on Monday, August 10, a 30-billion-parameter model you can download free from Hugging Face and run on a Mac or PC with one consumer graphics card. No API key. No monthly invoice. Zuckerberg tied the release to a broader pitch: he wants Washington to loosen restrictions on training data so American labs can keep pace with open-source AI coming out of China.\n\nThe timing says a lot.\n\nMuse Glimmer carries an Apache 2.0 license, meaning developers can modify and redistribute it with almost no strings attached. It's built for agentic work: running local agents, calling functions, writing and debugging code, and acting as a judge for other models' outputs. According to Meta AI Research's own announcement, the company tested it against SWE-Bench, DeepSearch QA, MCP-Atlas and tau-Bench, and says it holds its own against similarly sized open models like Google's Gemma 4-31B and Alibaba's Qwen 3.6-27B. Alexandr Wang, Meta's chief AI officer, posted on X that open weights for a version of Muse Spark 1.2 are coming soon too, calling Glimmer and Spark two very different models, both headed into people's hands.\n\nFor a startup building an AI agent, that difference matters right away. Instead of metering every call through Anthropic's or OpenAI's API and watching the bill climb with usage, a small team can load Muse Glimmer onto a single consumer GPU, wire up their own scaffolding, and never see a per-token charge again. That's the pitch Meta AI Research leans on in its own blog post announcing the model: local agents, offline coding assistants, and evaluation pipelines that never have to phone home to someone else's data center.\n\nFour months ago, Meta did the opposite. On April 8, Meta Superintelligence Labs, the unit Wang has run since Meta poured billions into Scale AI last year, shipped its first model, Muse Spark, and kept it closed. Wang said at the time that early training runs had triggered high-risk signals, particularly around bio-risk, and that is why the weights stayed locked up. It was a real reversal. For five years Zuckerberg had positioned Llama as the open alternative to OpenAI and Google. Muse Spark ended that streak.\n\nIt didn't land well. Bloomberg reported that Muse Spark's early benchmark results trailed rivals at launch, and CNBC reported that investors pushed Zuckerberg for a clearer AI strategy once the results came in. Meta pressed ahead anyway. Muse Spark 1.1 opened as a paid API on July 9, priced at $1.25 per million input tokens and $4.25 per million output tokens, roughly a quarter of what OpenAI and Anthropic charge for comparable access. Then on August 5, according to VentureBeat, Meta rolled out Muse Code, a coding agent built to compete directly with Anthropic and OpenAI's own coding tools.\n\nMuse Glimmer looks like the other half of that bet. Businesses have grown wary of ballooning AI bills, and recent cybersecurity incidents tied to models from Anthropic, OpenAI and Meta itself have made buyers cautious about handing everything to a closed API, as Rappler reported alongside Monday's launch. Meta wants that caution to work in its favor.\n\nFrankly, this isn't a rebuke of closed AI labs so much as Meta hedging both sides of the table. It still wants the subscription-and-API business that Sam Altman and Dario Amodei have built at OpenAI and Anthropic, which is why Muse Spark stays closed and priced. And it wants the developer goodwill that made Llama the default starting point for open-source builders for years, which is why Glimmer ships free with permissive terms attached.\n\nWhether that goodwill still exists is the real question. Developers who watched Meta wall off Muse Spark in April won't forget it just because Glimmer showed up in August. What they will notice is that a competent 30-billion-parameter model now runs on hardware they already own, for nothing, while OpenAI and Anthropic keep charging by the token.\n\n**Also read:** [SpaceX Is Closing In On A $6 Billion Deal For Israeli AI Startup Decart](https://startupfortune.com/spacex-is-closing-in-on-a-6-billion-deal-for-israeli-ai-startup-decart/) • [ByteDance's Zhang Yiming Bans AI Model Distillation Even as Rivals Race Ahead](https://startupfortune.com/bytedances-zhang-yiming-bans-ai-model-distillation-even-as-rivals-race-ahead/) • [TSMC's July sales jumped 45% as AI chip demand keeps outrunning tariffs](https://startupfortune.com/tsmcs-july-sales-jumped-45-as-ai-chip-demand-keeps-outrunning-tariffs/)", "url": "https://wpnews.pro/news/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop", "canonical_source": "https://startupfortune.com/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop/", "published_at": "2026-08-10 10:39:54+00:00", "updated_at": "2026-08-10 10:45:53.881386+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-policy"], "entities": ["Meta", "Muse Glimmer", "Mark Zuckerberg", "Hugging Face", "Alexandr Wang", "Google", "Alibaba", "Apache 2.0"], "alternates": {"html": "https://wpnews.pro/news/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop", "markdown": "https://wpnews.pro/news/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop.md", "text": "https://wpnews.pro/news/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop.txt", "jsonld": "https://wpnews.pro/news/meta-releases-muse-glimmer-an-open-weight-ai-model-that-runs-on-a-laptop.jsonld"}}