Is Muse Glimmer open source?

Yes, in the practical sense: Muse Glimmer ships under Apache 2.0, a permissive OSI-approved license with no revenue-sharing strings — unlike Qwen3.8-Max or Kimi K3, which attach commercial revenue-share terms. That makes Glimmer genuinely open for commercial use.

What is Muse Glimmer good for?

At 30B dense and multimodal, Muse Glimmer targets local, always-on agent workflows that run on consumer hardware — the opposite end of the spectrum from trillion-parameter frontier models. It is a value/self-host pick, not a top-of-leaderboard accuracy pick.

Quick Answer

What Is Meta Muse Glimmer? 30B Open Model (Aug 2026)

Q: What is Meta Muse Glimmer?

Muse Glimmer is Meta's 30-billion-parameter dense, multimodal, agent-focused model, released August 10, 2026 under an Apache 2.0 license. It is optimized for always-on local agents and consumer hardware, marking Meta's re-entry into the open-weight frontier after the Llama-to-Muse pivot.

Published: August 12, 2026

The Short Answer

Muse Glimmer is Meta’s 30-billion-parameter dense, multimodal, agent-focused open-weight model, released August 10, 2026 under a permissive Apache 2.0 license. It’s built for always-on local agents on consumer hardware — Meta’s clearest re-entry into the open-weight frontier since the Llama-to-Muse pivot.

What Changed

After pivoting its flagship line to the closed Muse Spark family, Meta reopened the open-weight door with Glimmer. The pitch is local-first agents: a dense 30B model small enough to run on a workstation, multimodal, and licensed so businesses can ship it without negotiating terms.

Key Facts (as of August 2026)

	Muse Glimmer
Vendor	Meta
Size	30B dense (not MoE)
Modality	Multimodal
License	Apache 2.0 (permissive, no revenue share)
Released	August 10, 2026
Best for	Always-on local agents, consumer hardware

Why It Matters

The license is the headline. Apache 2.0 with no revenue-share stands out against the 2026 open-weight trend of commercial strings: Qwen3.8-Max and Moonshot’s Kimi K3 both attach revenue-share terms for large commercial users. Glimmer gives self-hosters and startups a genuinely unencumbered 30B multimodal agent base.

Which Should You Pick?

Truly permissive local agent base → Muse Glimmer (Apache 2.0).
Maximum open-weight capability → Qwen3.8-Max or Kimi K3 (bigger, but with revenue-share strings).
Smallest footprint on a laptop → Glimmer’s 30B dense design fits consumer GPUs.

The Reality Check

At 30B dense, Glimmer won’t top frontier leaderboards against trillion-parameter models. Its value is the license + local footprint combination: an agent brain you can embed, ship, and run offline without legal friction.

Sources

AINews — Meta re-enters open-weight frontier with Muse Glimmer: news.smol.ai
Implicator — Meta opens Muse Glimmer: implicator.ai
AI Release Tracker — Muse Glimmer (Meta, Aug 10, 2026): aireleasetracker.com