- Sources: Meta AI Research blog, Muse-Glimmer-30B on Hugging Face, USAGE_POLICY.md, discussion
- Summary: Meta Superintelligence Labs published Muse Glimmer on 2026-08-10 on the Meta AI Research blog, and the model card lists Meta Superintelligence Lab as the author. The Hugging Face repository meta-models/Muse-Glimmer-30B was created 2026-08-09 at 17:51 UTC, carries the license identifier apache-2.0, and reports 29,776,626,688 BF16 parameters. Meta states a quantized build runs on a single 24 GB consumer GPU. The repository also ships a USAGE_POLICY.md next to the Apache-2.0 license text. The policy prohibits military, warfare, nuclear, espionage, and ITAR-controlled applications, and the operation of critical infrastructure, transportation technologies, and heavy machinery. It separately prohibits engaging in or facilitating any action to intentionally circumvent or remove usage restrictions or other safety measures, or to enable functionality disabled by Meta. The policy also states that Muse Glimmer is not intended for individuals under the age of 18. Apache-2.0 carries no field-of-use restriction, so the two documents do not state the same terms, and the deployable scope is narrower than the license identifier alone shows. A team evaluating the model for a regulated or safety-critical setting should read the usage policy rather than the license. This digest does not assess whether the policy is enforceable alongside the Apache-2.0 grant, because no source addresses that question. The benchmark comparisons against Gemma4-31B and Qwen3.6-27B are vendor figures with no independent reproduction and are not restated here.
- Why it matters: A model this size under a permissive license identifier, quantized to fit a single 24 GB consumer GPU, puts agent workloads that needed a hosted API onto a developer workstation.
send feedback on this story