Muse Glimmer 30B

Name: Muse Glimmer 30B
Price: 0.525 USD
Author: Meta

Meta · Muse · released 2026-08-09 · data checked 2026-11-01

Muse Glimmer 30B is a 29.6B-parameter dense model from Meta (United States), released 2026-08-09 under the Apache-2.0 license. It scores 72.6 on the Small LM Index, #5 of 34 medium models, with 2 independently measured benchmarks. It needs about 19.2 GB of memory at 4-bit with an 8K-token context and costs $0.53 per million tokens through the cheapest API provider.

Meta's first open-weight model since Llama 4 and the first under Apache 2.0. Runs on a single consumer GPU at 4-bit (~17 GB).

Small LM Index

72.6

#5 of 34 scored medium models

Likely 71.3–74.0 · from 4 of 6 core benchmarks

Parameters

29.6B

Dense

Price per 1M tokens

$0.53

$0.30 in / $1.20 out · blended 3:1

Context window

131K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 5 reported by Meta, 2 independently measured. How we label sources.

GPQA Diamond Independent Knowledge · Measured by Artificial Analysis83.5

Class best: Nanbeige4.3 33B 89.5

Humanity's Last Exam Independent Reasoning · HLE text, measured by Artificial Analysis22.0

Class best: Qwen3.8 27B 30.8

AIME Lab-reported Math · AIME 202694.7

Class best: Ternary Bonsai 2 27B 95.8

SWE-bench Verified Lab-reported Coding76.0

Class best: Qwen3.6 27B 77.2

Terminal-Bench Lab-reported Agentic · Terminal-Bench 2.1 (terminus2)51.7

Class best: Nanbeige4.3 33B 80.9

IFBench Lab-reported Instruction following77.0

Class best: Sage-A1.5 81.6

MMMU-Pro Lab-reported Vision74.0

Class best: Gemma 4 31B 76.9

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 86

LiveCodeBench≈ 83

Specifications

Lab: Meta (US · United States)
Architecture: Dense transformer
Total parameters: 29.6B
Active parameters: All (dense)
Size class: Medium
Context window: 131K tokens
Input modalities: text, image
Reasoning: Reasoning (thinking)
License: Apache-2.0
Openness: Open weights
Memory to run: ~19.2 GB at 4-bit, 8K context (more sizes)
Hugging Face: 189.7K downloads (30 days) · 2.1K likes
API providers: 1 on OpenRouter
API price: $0.30 in / $1.20 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Muse Glimmer 30B vs Granite 4.2 30B (30B, index 67.0)
Muse Glimmer 30B vs Qwen3 30B-A3B Thinking 2507 (30.5B, index 59.3)
Muse Glimmer 30B vs Gemma 4 31B (30.7B, index 70.1)
Muse Glimmer 30B vs Kanana-2 30B-A3B Thinking (30.7B, index 50.7)

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses meta-models/Muse-Glimmer-30B-GGUF (Q4_K_M). Swap the tag for another size: Q4_K_M, Q4_K_XL.

Ollama

ollama run hf.co/meta-models/Muse-Glimmer-30B-GGUF:Q4_K_M

llama.cpp

llama-server -hf meta-models/Muse-Glimmer-30B-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/Muse-Glimmer-30B-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve meta-models/Muse-Glimmer-30B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	131K context
16-bit F16	60 GB	61 GB	62 GB
8-bit Q8_0	33 GB	33 GB	34 GB
4-bit Q4_K_M	19 GB	20 GB	21 GB

Check it against your GPU, Mac or phone →