Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local
- Source
- x.com
- Date
Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs. In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license. 🧵👇
Context
Muse Glimmer is a 30-billion-parameter model from Meta's Superintelligence Labs, released with its weights under the Apache 2.0 license, which lets anyone download, modify and run it. Meta announced it as a model for "local, always-on workflows", meaning an assistant that runs on the user's own computer and calls tools over long tasks instead of sending everything to a cloud service.
The claim that it runs on consumer hardware depends on compression. According to Meta's release post, a 30-billion-parameter model at full precision would need over 55 GB of memory. Meta quantizes the weights, storing each number at roughly 4-bit precision, which shrinks the to under 20 GB. Meta says that leaves room for the model and its companion components, including a speculative-decoding drafter used to speed up generation, to run together within a 24 GB or 32 GB memory budget. Meta's description of a Mac or a PC with a single consumer GPU therefore implies a machine with that much memory available, not any recent computer. Meta reports speed measurements on MacBook M4 Max and M5 Max machines and an RTX 5090 graphics card. It also says the model was trained by logit on the outputs of Muse Spark, a much larger teacher model, a recipe Meta says transfers the larger model's agentic reasoning.
The performance claim is weaker evidence. Meta's post says Muse Glimmer "performs strongly for its size class" against Gemma4-31B and Qwen3.6-27B, two models of similar size, but the comparison figures sit in a chart image and a linked technical report that this explanation did not review. Until independent tests appear, the 30-billion-parameter size, the open license and the roughly 20 GB footprint are the facts to rely on, and the benchmark standing is Meta's own account.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
Checking sign-in…
Loading comments…