Muse-Glimmer-30B

Muse family · 30.0B params

13recorded runs

161.9best tok/s

11verified

Evidence: 11 sourced runs · 2 awaiting review · 1 independent source · first recorded Aug 10, 2026 · last recorded Aug 29, 2026. What these labels mean.

What the Muse-Glimmer-30B numbers mean

The fastest verified result on our board is 161.9 output tok/s on an Intel Arc Pro B70 x4 at UD-Q8_K_XL with a 32768-token context. That is about 6.2 ms per generated token — 500 tokens takes roughly 3.1 seconds. Across the 13 records we hold for this model the range is 6.0–161.9 tok/s on 10 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does Muse-Glimmer-30B need?

Muse-Glimmer-30B has 30.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	20.2 GB	24 GB
Q8_0 (8-bit)	37.5 GB	48 GB
FP16 (16-bit)	69.8 GB	96 GB

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

Models with a similar footprint

Model	Params	Q4_K_M weights
Hy-MT2-30B-A3B	30B	16.9 GB
Nemotron-3-Nano-Omni-30B-A3B-Reasoning	30B	16.9 GB
North-Mini-Code-1.0	30B	16.9 GB
Qwen3-Coder-30B-A3B-Instruct	30B	16.9 GB
granite-4.2-30b	29B	16.3 GB
QUEST-30B-RL-B70-Turbo	31B	17.4 GB

Ranked by distance from Muse-Glimmer-30B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Intel Arc Pro B70 x4 · 32 GB	Muse-Glimmer-30B	UD-Q8_K_XL	llama.cpp 10433; 030ebb558 + Muse TP4/WOQ patc…	32768	161.9 ✓	—	—	src
RTX PRO 6000 Blackwell · 96 GB	Muse-Glimmer-30B	NVFP4	vllm 0.1.dev1+g075d645af	32768	152.2 ✓	4396.9	—	src
Radeon AI Pro R9700 · 32 GB	Muse-Glimmer-30B	Dynamic-Q4_K_XL + DFlash2-Q4_K_M	llama.cpp 0.3.0-dev / build 10680 / commit d7b…	32768	134.6 ✓	380.6	23.6	src
RTX 5090 · 32 GB	Muse-Glimmer-30B	NVFP4	vllm 0.1.dev1+g075d645af	4096	124.8 ✓	2986.4	—	src
RTX PRO 6000 Blackwell · 96 GB	Muse-Glimmer-30B	FP8	vllm 0.1.dev1+g075d645af	32768	106.6 ✓	7584.4	—	src
RTX 3060 x2 · 12 GB	Muse-Glimmer-30B	UD-Q4_K_XL	llama.cpp b10443	65536	56.0 ✓	634.6	18.7	src
RTX PRO 6000 Blackwell · 96 GB	Muse-Glimmer-30B	Q8_K_XL	llama.cpp 10355	131072	48.1 ✓	3624.0	32.8	src
Radeon AI Pro R9700 x3 · 32 GB	Muse-Glimmer-30B	UD-Q8_K_XL	llama.cpp b10434	834	47.8 ✓	787.5	—	src
Intel Arc Pro B70 · 32 GB	Muse-Glimmer-30B	Q4_K_XL	llama.cpp b284 SYCL (d2f83055d) GGML_SYCL_F16=…	131072	29.2 ✓	1301.0	—	src
RTX PRO 6000 Blackwell · 96 GB	Muse-Glimmer-30B	BF16	llama.cpp 10355	131072	28.8 ✓	3665.0	52.3	src
RTX 3080 · 10 GB	Muse-Glimmer-30B	Q4_K_M	llama.cpp b10472	16384	6.0 ✓	672.4	8.8	src
RTX 4060 Ti x2 · 16 GB	Muse-Glimmer-30B	UD-Q6_K_XL	llama.cpp —	125000	44.0 …	841.2	—	src
RTX 3090 x2 · 24 GB	Muse-Glimmer-30B	Unsloth-Dynamic-Q6_K_XL	llama.cpp —	2048	31.7 …	1604.0	—	src