vs
0:6
overall capability wins (0 ties across 6 scored). Muse Glimmer leads overall.
- Reasoning0:3
- Arena0:1
- Specs0:2
Verdict
Muse Glimmer edges reasoning & knowledge; Muse Glimmer ranks higher on LMArena (+261 Elo); Muse Glimmer offers larger context (131K vs 128K).
- Muse Glimmer edges reasoning & knowledge
- Muse Glimmer ranks higher on LMArena (+261 Elo)
- Muse Glimmer offers larger context (131K vs 128K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Llama 3.2 3B
- $0.04
- Muse Glimmer
- -
Price per 1M tokens
Full comparison
Organization
Meta
Meta
Family
Llama / Muse
Llama / Muse
License
Llama 3.2
Apache 2.0
Open weights
Yes
Yes
Release
Sep 25, 2024
Aug 10, 2026
Knowledge cutoff
-
2026-01-04
API / provider
Together / Fireworks (ref.)
No primary API price listed
Modalities
text → text
text, image → text
Specs
Context window
128K
131K
Max output
8K
-
Parameters
3B
30B
Pricing
Input $/1M
$0.03
—
Output $/1M
$0.05
—
Blended $/1M (3∶1)
$0.04
-
Speed
tok/s
220
-
TTFT (s)
0.1
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
25.5
83.5
Humanity's Last Exam
5.4
22.0
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
27.0
AA-LCR v1.1
4.3
83.3
CritPt
—
2.6
MMMU-Pro
—
74.3
IFBench
26.2
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
76.0
SWE-bench Pro
—
51.2
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
51.7
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
23.4
—
SciCode
—
44.9
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,360
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,167
1,427
| Metric | Llama 3.2 3B | Muse Glimmer | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Meta | Meta | — |
| Family | Llama / Muse | Llama / Muse | — |
| License | Llama 3.2 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 25, 2024 | Aug 10, 2026 | — |
| Knowledge cutoff | - | 2026-01-04 | — |
| API / provider | Together / Fireworks (ref.) | No primary API price listed | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 131K | B +3K |
| Max output | 8K | - | — |
| Parameters | 3B | 30B | B +27 |
| Pricing | |||
| Input $/1M | $0.03 | — | — |
| Output $/1M | $0.05 | — | — |
| Blended $/1M (3∶1) | $0.04 | - | — |
| Speed | |||
| tok/s | 220 | - | — |
| TTFT (s) | 0.1 | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 25.5 | 83.5 | B +58.0 pts |
| Humanity's Last Exam | 5.4 | 22.0 | B +16.6 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 27.0 | — |
| AA-LCR v1.1 | 4.3 | 83.3 | B +79.0 pts |
| CritPt | — | 2.6 | — |
| MMMU-Pro | — | 74.3 | — |
| IFBench | 26.2 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 76.0 | — |
| SWE-bench Pro | — | 51.2 | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 51.7 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | 23.4 | — | — |
| SciCode | — | 44.9 | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,360 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,167 | 1,427 | B +261 Elo |
Llama 3.2 3B
- Context
- 128K / 8K out
- Parameters
- 3B
- Price
- $0.03 / $0.05
- Speed
- 220 tok/s · 0.1s TTFT
- Modalities
- text → text
- License
- Llama 3.2
Edge-oriented Llama 3.2 text model with 128K context for on-device assistants.
Muse Glimmer
- Context
- 131K
- Parameters
- 30B
- Price
- —
- Speed
- —
- Modalities
- text, image → text
- License
- Apache 2.0
Meta's first open-weight model from Superintelligence Labs — 30B dense multimodal agentic model distilled from Muse Spark; Apache 2.0 with 128k context, built to run local agents on a single consumer GPU.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.