DeepSeek-V4-Flash

DeepSeek · 158.1B (active 13B) · MoE · ctx 1000K · MIT · HuggingFace

DeepSeek-V4-Flash (DeepSeek) is a 158.1B parameters MoE (13B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 90.55 GB VRAM for Q4 (weights 86.95 GB + KV 2.6 GB + 1 GB overhead) and 319.8 GB for FP16. About 8 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 391.6 tok/s (estimated); MI300X 192GB ≈ 259.4 tok/s (estimated); H200 141GB SXM5 ≈ 235.0 tok/s (estimated); Gaudi 3 128GB ≈ 160.4 tok/s (estimated); Mac Studio M3 Ultra 96GB ≈ 40.1 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

31 hardware configs — sorted by tok/s. Default fit context 8K.