vram.run Models Hardware Providers Cloud State of Inference
API provider data is live · Hardware & cloud pricing curated 2026-02-23

DeepSeek-V4-Flash

by deepseek-ai

158B params · text-generation · 1.5k likes · 2.4M downloads

DeepSeek-V4-Flash is a 158B parameter model. At Q4 quantization it requires 79GB of VRAM. It runs comfortably on H200 SXM 141 GB (44 tok/s).

Inference providers

Provider$/1M in$/1M outThroughput
Novita86 tok/s
Fireworks59 tok/s
Featherless
deepinfra13 tok/s

GPU compatibility

GPUVRAMQ4 DecodeVerdict
M4 Max 128GB128GB3 tok/stight
H200 SXM 141 GB141GB44 tok/scomfortable
M2 Ultra 192GB192GB4 tok/stight
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0