GLM-5.3-Flash

GLM family · 321.0B params

21recorded runs

264.6best tok/s

20verified

Evidence: 20 sourced runs · 1 awaiting review · 2 independent sources · first recorded Aug 27, 2026 · last recorded Oct 13, 2026. What these labels mean.

What the GLM-5.3-Flash numbers mean

The fastest verified result on our board is 264.6 output tok/s on an CMP 170HX x4 at W4A16 with a 262144-token context. That is about 3.8 ms per generated token — 500 tokens takes roughly 1.9 seconds. Across the 21 records we hold for this model the range is 17.4–264.6 tok/s on 10 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does GLM-5.3-Flash need?

GLM-5.3-Flash has 321.0B 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)	208.4 GB	multi-GPU or CPU offload
Q8_0 (8-bit)	393.0 GB	multi-GPU or CPU offload
FP16 (16-bit)	739.1 GB	multi-GPU or CPU offload

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
DeepSeek-V4-Flash-Vision-Exp	305B	171.6 GB
keys-DeepSeekV4Flash-Vision-EXP-ablit	305B	171.6 GB
DeepSeek-V4-Flash-0731	304B	171.0 GB
DeepSeek-V4-Flash-0731-ROCmFP3	304B	171.0 GB
Hy3	299B	168.2 GB
DeepSeek-V4-Flash	291B	163.7 GB

Ranked by distance from GLM-5.3-Flash's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
CMP 170HX x4 · 64 GB	GLM-5.3-Flash	W4A16	vllm 0ed7d3e7f3	262144	264.6 ✓	—	—	src
RTX PRO 6000 Blackwell x2 · 192 GB	GLM-5.3-Flash	EXL3-3.0bpw	vllm 0.1.dev20111+g7f1e92bec.d20260827	700000	159.3 ✓	2396.3	186.6	src
RTX PRO 6000 Blackwell x2 · 192 GB	GLM-5.3-Flash	exl3	vllm 0.1.dev20111+g7f1e92bec.d20260827	131072	128.6 ✓	371.3	—	src
DGX Spark x4 · 128 GB	GLM-5.3-Flash	EXL3 4bpw	tensorfold v0.6.0	500000	124.0 ✓	—	—	src
DGX Spark x4 · 128 GB	GLM-5.3-Flash	NVFP4	vllm knapcio stack	500000	115.0 ✓	—	—	src
DGX Spark x4 · 128 GB	GLM-5.3-Flash	EXL3 4bpw	vllm knapcio stack	500000	115.0 ✓	—	—	src
Mac Studio M3 Ultra	GLM-5.3-Flash	4-bit	tensorfold 0.6.2	256	78.0 ✓	—	—	src
RTX 3090 x8 · 24 GB	GLM-5.3-Flash	AWQ-4bit	vllm 0.28.1rc1.dev125+g85872952a.glm53sm8…	204800	75.2 ✓	1020.0	183.4	src
DGX Spark · 128 GB	GLM-5.3-Flash	EXL3 2.05 bpw	vllm sha256:905c02933be6021301db2dc284e24…	262144	64.1 ✓	—	—	src
DGX Spark · 128 GB	GLM-5.3-Flash	EXL3 2.05 bpw	vllm sha256:905c02933be6021301db2dc284e24…	512	64.1 ✓	—	—	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	NVFP4	vllm 0.28.1rc1.dev580	126720	46.3 ✓	—	—	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	W4A16 AutoRound	vllm v0.31.0	400000	43.1 ✓	—	—	src
RTX 3090 x8 · 24 GB	GLM-5.3-Flash	Unsloth-Dynamic-IQ4_XS	llama.cpp b10667-2e0e57f1	131072	42.1 ✓	148.4	165.9	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	W4A16 AutoRound	vllm v0.31.0	1000	39.4 ✓	—	—	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	W4A16 AutoRound	vllm v0.31.0	128000	38.5 ✓	—	—	src
DGX Spark · 128 GB	GLM-5.3-Flash	EXL3 2.05 bpw	vllm sha256:905c02933be6021301db2dc284e24…	40000	35.5 ✓	—	—	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	NVFP4	vllm glm53-sm121-v13	327680	30.0 ✓	—	—	src
RTX 3090 · 24 GB	GLM-5.3-Flash	EXL3 3.05 bpw	exllamav2 v1.5.1	131072	28.1 ✓	—	—	src
DGX Spark x2 · 128 GB	GLM-5.3-Flash	EXL3 4-bit	vllm Entrpi/vllm-glm-5.3-flash-spark	3700	27.4 ✓	1369.0	—	src
Radeon AI Pro R9700 · 32 GB	GLM-5.3-Flash	UD-IQ1_M	llama.cpp 2e0e57f10	32768	17.4 ✓	39.6	92.2	src
Radeon AI Pro R9700 x3 · 32 GB	GLM-5.3-Flash	compressed-tensors	sglang —	12000	250.6 …	3482.2	—	src