ATRIUMsearch → argument graph
Article · 2026-08-10 · 5 moments

Introducing Muse Glimmer

Introducing Muse Glimmer Meta are back in the open weights game! Muse Glimmer is a brand new 30B model under a clean Apache 2.0 license (a step up from the janky Llama licenses of old). They claim to have optimized it for exactly the kind of things I'm looking for in a local model: End-to-end Agentic Task Completion. Muse Glimmer achieves strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, 𝛕-Bench and SWE-Bench, which measure its ability to work within scaffolds, ✦ AI generated

Read original ↗
HOST Simon Willison
01
Context◆

I really like this size of model because, on a machine with 32 GB or more of RAM (mine has 128 GB), it leaves plenty of space for running other applications at the same time.

The author endorses this model size for its modest footprint: on machines with 32 GB+ RAM, including his 128 GB setup, it leaves ample room to run other apps concurrently.

transcript

Simon Willison: I really like this size of model, because if a machine has 32 GB of RAM or more (mine has 128GB) it leaves plenty of space for running other applications at the same time.

02
Claim◆

Meta are back in the open weights game, releasing Muse Glimmer as a brand new 30B model under a clean Apache 2.0 license, which is a step up from the janky Llama licenses of old.

Meta returns to the open-weights space with Muse Glimmer, a new 30B model released under a clean Apache 2.0 license, contrasted favorably against prior Llama licensing.

transcript

Simon Willison: Meta are back in the open weights game! Muse Glimmer is a brand new 30B model under a clean Apache 2.0 license (a step up from the janky Llama licenses of old).

04
Data

Muse Glimmer is optimized for end-to-end agentic task completion, achieving strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, τ-Bench and SWE-Bench.

The model is said to be tuned for end-to-end agentic work, scoring well on benchmarks that require working within scaffolds, writing and debugging code, and resolving multi-turn requests.

transcript

Simon Willison: End-to-end Agentic Task Completion. Muse Glimmer achieves strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, τ-Bench and SWE-Bench, which measure its ability to work within scaffolds, write and debug code, and resolve multi-turn requests from start to finish.

provides context · 1

Highlight slides
Meta returns to open weights with Muse Glimmer✦ from: Meta are back in the open weights game, releasing Muse Glimmer as a brand new 30B model under a clean Apache 2.0 license, which is a step up from the janky Llama licenses of old.Modest Model Footprint✦ from: I really like this size of model because, on a machine with 32 GB or more of RAM (mine has 128 GB), it leaves plenty of space for running other applications at the same time.
Related episodes
Article · Interconnects

Kimi K3: The open-weights escalation

→
Article · Import AI

Import AI 464: Fables writes GPU kernels; AI automation; and analog computation

→
Audio · We Study Billionaires

TIP825: Meta, Adobe, Booking Holdings w/ Stig Brodersen, Tobias Carlisle & Hari Ramachandra

→