Muse Glimmer is an open-weight agentic model released by Meta on August 10, 2026, described as sized to run on a Mac or PC with a single graphics card. It was distilled from Muse Spark by Meta Superintelligence Labs, the division led by former Scale AI chief executive Alexandr Wang. Meta said the model "performs strongly for its size class" on several benchmarks, without naming the benchmarks or the comparison set (Source: reuters.com; businessinsider.com).
The release marks a partial return to weight release after the closed-weight Muse Spark line, which had been a break from Meta's Llama practice. Meta said alongside the Glimmer release that it plans to release the weights of Muse Spark 1.2, described as its most advanced model; as of August 10, 2026 that is an announced intention rather than a completed release.
Capabilities as stated
Meta specified the model at 30 billion dense parameters, distilled from Muse Spark 1.2 (Source: spyglass.org). Beyond that it has published no benchmark table, training-data description, or evaluation report for Glimmer. The remaining claims on the record are that the model is agentic, that it runs on a single consumer graphics card, and that it "performs strongly for its size class." The absence of a published model card distinguishes the release from Muse Spark, which shipped with a 158-page safety-focused report.
Release context
Glimmer was released with an essay of roughly 6,400 words by Mark Zuckerberg titled "The Future is for Everyone" (The Future is for Everyone (Zuckerberg, August 2026)), which argues for reducing U.S. regulatory friction on open-weight development. Zuckerberg wrote that "the notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic," called it "surprising" that discourse from many AI developers was "filled with doom," and argued that "foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data" and that "U.S. policy must reduce this additional friction if we want American open source models to lead over time." He named data-use and distillation rules as the specific friction and rejected restrictions on access to foreign open-source models. See Open-Weight Frontier Models and AI Power Concentration.
M.G. Siegler wrote that the accompanying essay repeats verbatim at many points an op-ed Zuckerberg had published in the Wall Street Journal a couple of weeks earlier, that Glimmer sits below even Meta's own current line-up rather than at the frontier, and that the open question is whether Meta will release weights for its unreleased frontier model, codenamed Watermelon (Source: spyglass.org).
Meta's plan to release an open-weight version of Muse Spark 1.2 became public again on August 11, 2026 through a briefing summary for which no article body could be retrieved; the timing and scope of that release are unverified beyond the summary (Source: theinformation.com).
Meta announced two further items with the release: a $1 billion fund to support communities near its data centres, against company AI-infrastructure spending set to reach as much as $145 billion in 2026; and that it will give its independent directors power to approve the safety criteria for model releases. Meta shares, down about 10% year to date, rose nearly 3% in premarket trading on the announcement (Source: reuters.com).
Relationships
- instance-of: Open-Weight Frontier Models
- depends-on: Muse Spark (Meta Superintelligence Labs) — Glimmer is distilled from it
- deployed-by: Meta AI
- related: Mark Zuckerberg, Alexandr Wang, AI Power Concentration, gpt-oss (OpenAI open-weight models)
Open questions
Whether Meta releases the Muse Spark 1.2 weights as announced, and on what licence, is not yet resolved. No benchmark figures, parameter count, or safety evaluation for Glimmer has been published, so its position relative to other open-weight models cannot be stated.