Muse Glimmer

Muse Glimmer ranks first on AA-LCR (Artificial Analysis Long Context Reasoning).

Meta Released 10 Aug 2026

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Meta Aug 2026

Muse Glimmer

Open weights

No Epoch Capabilities Index score yet

IFBench 77% #15 /70
MMMU-Pro 74% #36 /72
SciCode 43.6% #23 /47

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
AA-LCR (Artificial Analysis Long Context Reasoning)	80%	#1 of 31	Model card
IFBench	77%	#15 of 70	Model card
LMArena (Elo)	1,425	#98 of 333	Official leaderboard
SEA-HELM	72.1	#26 of 62	Official leaderboard
MMLU-ProX	80.1%	#9 of 17	Model card
Humanity's Last Exam	22%	#94 of 147	Model card
BEAM	65.1%	#1 of 1	Model card
Belebele	87.8%	#1 of 1	Model card
BOUQuET	36.5	#1 of 1	Model card
FLORES+	47.5	#1 of 1	Model card
Linguini	8.80%	#1 of 1	Model card
Omnilingual-Gaia2	60.4%	#1 of 1	Model card

Math and science

Benchmark	Score	Rank	Source
GPQA Diamond	83.5%	#87 of 284	Model card
DTBench	76%	#68 of 178	Official leaderboard
AIME 2026	94.7%	#14 of 34	Model card
CritPt	2.57%	#82 of 160	Epoch AI
ARC-AGI	66.9%	#13 of 19	Model card

Coding

Benchmark	Score	Rank	Source
SWE-bench Verified	76%	#41 of 150	Model card
SciCode	43.6%	#23 of 47	Model card
Code Arena	1,355	#98 of 124	Official leaderboard
SWE-Bench Pro	51.2%	#64 of 77	Model card

Agents and tools

Benchmark	Score	Rank	Source
PinchBench	90.5%	#4 of 13	Model card
ScreenSpot-v2	93.3%	#5 of 14	Model card
ScreenSpot-Pro	75.4%	#9 of 25	Model card
LLM Chess	214	#51 of 124	Official leaderboard
MCP Atlas	75.5%	#20 of 40	Model card
τ³-bench Banking	23.5%	#10 of 20	Model card
OSWorld-Verified	65.9%	#25 of 37	Model card
WildClawBench	47.6%	#4 of 5	Model card
SkillsBench	44.3%	#11 of 13	Model card
DeepSearchQA	74.6%	#19 of 22	Model card
Terminal-Bench 2.1	51.7%	#47 of 53	Model card
WitnessBench	0.21%	#11 of 12	Official leaderboard
Meta Agents Research Environments (ARE) / Gaia2	43.3%	#3 of 3	Model card

Professional work

Benchmark	Score		Rank	Source
GDPval-AA v2	953		#25 of 26	Model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
OmniDocBench v1.5	75.8	#4 of 13	Model card
MMMU-Pro	74%	#36 of 72	Model card
CharXiv	78.8%	#36 of 67	Model card

Safety and security

Benchmark	Score	Rank	Source
CIMemories	26.4%	#1 of 2	Model card
AgentDojo	28.4%	#7 of 7	Model card
PrivacyLens	49.2%	#1 of 1	Model card

Other

Benchmark	Score		Rank	Source
ACCoRD	72.0		#35 of 150	Official leaderboard

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device (Meta) and model card 10 Aug 2026

MCP Atlas (Public) 75.5
DeepSearch QA 74.6
τ3-Banking 23.5
WildClawBench 47.6
GDPVal-AA v2 953
Gaia2 43.3
SkillsBench (with skills) 44.3
OSWorld-Verified 65.9
SWE-Bench Pro 51.2
SWE-Bench Verified 76
TerminalBench 2.1 (with terminus2) 51.7
SciCode 43.6
Charxiv Reasoning 78.8
ScreenSpot Pro 75.4
OmniDocBench v1.5 75.8
MMMU Pro 74
CI Memories Violation 26.4 (lower is better), Coverage 64.8 26.4
Siren AgentDojo Attack Success Rate 28.4 (lower is better), Utility 94.2 28.4
IFBench 77
AIME 2026 94.7
GPQA Diamond (AA) 83.5
HLE Text (AA) 22
AA-LCR 80
Beam128K 65.1

Its system card

Every result in the longer documents the lab published with it. 13 of them aren't in the launch post.

Muse Glimmer 30B: An On-Device Agentic Model 19 Oct 2026