Muse Glimmer 30B
Meta Superintelligence LabsAug 10, 2026N/A
NewOpen weights
Meta reopens its weights: a dense 30B multimodal agentic model designed to run on-device, framed as an open version of Spark. Two weeks old — treat early benchmark chatter with skepticism until indexers catch up.
- Status
- Generally available
- License
- Open weights · Custom
- Context window
- 256K
- Max output
- 32K
- Input · $/M
- —
- Output · $/M
- —
- Cached input · $/M
- —
- Blended · $/M
- —
- Parameters
- 30B dense
- Inputs
- Text + images
Benchmarks
This model has no published scores on the benchmarks we track yet. Specs and pricing below the fold are still fully sourced.
Reasoning & knowledge
GPQA Diamond
no public score
MMLU-Pro
no public score
Humanity's Last Exam
no public score
ARC-AGI-2
no public score
Coding
SWE-bench Verified
no public score
SWE-bench Pro
no public score
LiveCodeBench
no public score
Agents & computer use
Terminal-Bench 2
no public score
OSWorld 2.0
no public score
BrowseComp
no public score
Math
AIME 2026
no public score
Known strengths
- On-device footprint
- Open weights return for Meta
- Agentic-first design
Caveats
- Too new for settled independent scores