Meta Muse Spark
The model that ended the Llama era — Meta's first closed flagship, now partially reopening.
You're looking at an AI model record: who built it, what it can do, and the hardware and data behind it.
Muse Spark (announced April 8, 2026) is the first flagship from Meta Superintelligence Labs and a strategic reversal: unlike every Llama before it, it launched proprietary, with no open weights. It takes text, image and speech input with a 260K-token context; Muse Spark 1.2 (August 5, 2026) is a coding-focused update co-trained with the Muse Code terminal agent and extends context to 1M tokens. The open story then partially reversed again: Meta released the smaller open-weight Muse Glimmer (30B) on August 10, 2026 and Zuckerberg pledged to open Muse Spark 1.2's weights, with no confirmed date as of this record.
Specifications & key facts
Each figure is labeled by basis: vendor spec, vendor claim, reported, estimate, or official. "Not publicly disclosed" means exactly that — we never invent values.
Overview
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Developed by
Competes with
Related
Sources
Muse Spark (Wikipedia) ↗ · CNBC on Muse Glimmer (Aug 2026) ↗
Record reviewed as of 2026-08-31 — see Methodology.