DeepSeek-V4-Flash

Deepseek family · 291.0B params

1recorded run

53.1best tok/s

1verified

Evidence: 1 sourced run · 1 independent source · first recorded Aug 10, 2026. What these labels mean.

What the DeepSeek-V4-Flash numbers mean

The fastest verified result on our board is 53.1 output tok/s on an Radeon AI Pro R9700 x4 at MQ2R with a 2052-token context. That is about 18.8 ms per generated token — 500 tokens takes roughly 9.4 seconds. Prompt processing on the same record runs at 481.0 tok/s, so a 2,000-token prompt is read in about 4.2 seconds.

How much VRAM does DeepSeek-V4-Flash need?

DeepSeek-V4-Flash has 291.0B 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)	189.0 GB	multi-GPU or CPU offload
Q8_0 (8-bit)	356.4 GB	multi-GPU or CPU offload
FP16 (16-bit)	670.1 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)

Radeon AI Pro R9700 x453.1

Models with a similar footprint

Model	Params	Q4_K_M weights
Hy3	299B	168.2 GB
DeepSeek-V4-Flash-0731	304B	171.0 GB
DeepSeek-V4-Flash-0731-ROCmFP3	304B	171.0 GB
DeepSeek-V4-Flash-Vision-Exp	305B	171.6 GB
keys-DeepSeekV4Flash-Vision-EXP-ablit	305B	171.6 GB
GLM-5.3-Flash	321B	180.6 GB

Ranked by distance from DeepSeek-V4-Flash's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Radeon AI Pro R9700 x4 · 32 GB	DeepSeek-V4-Flash	MQ2R	hipfire 0.3.0+5ccef2df2719	2052	53.1 ✓	481.0	—	src