Model · DeepSeek

DeepSeek-V4-Flash

304B MoE · Q4_K · 144.4 GB · open weights · curated for general, coding, research · repo created 2026-07-31

Too much for any tracked machine: 150.0 GB at 8K context; the largest we track has 112.6 GB usable.

What the model needs: weights, context cache and runtime overhead at four reference contexts. The total is the model's own; what differs per machine is the usable memory it has to fit into — VRAM on a graphics card, unified memory on a Mac.

Cache shape from the model's config: 43 of 43 layers cache the full context with multi-head latent attention, one 576-dimensional latent per token at f16 in place of separate K and V.

All memory figures are estimates: measured quantized file size + computed context memory + runtime overhead, with a 12% safety margin on your hardware. Real usage varies with runtime version and settings.

Every tracked GPU and Mac at 8K and 32K context with a full-precision (f16) cache — the same engine as the finder and the GPU pages.

Runs on 0 of 32 tracked devices at 8K context, easily or with ordinary headroom.

GGUF is the quantized single-file format local runtimes load (Ollama, LM Studio, llama.cpp); our memory figures are measured from the linked GGUF file. The original repo holds full-precision safetensors — a much larger download meant for GPUs with far more memory.

More from DeepSeek

• DeepSeek-R1about the same ◆
• DeepSeek-V4-Promore popular ▲

Every tracked model →