local.aiEarly access
HardwareModelsEnergyWallReferralsContact

Search

Find people, models, hardware, or pages

Account
HardwareModelsEnergyWallReferralsContactClaim your nameAccount
Model

DeepSeek V4 Flash

Intelligence rank
#21/238
70
Speed rank
#94/238
8m 24s
Fastest hardware: MacBook Pro M5 Max 128GB · 40-core GPU.

Quick Start

Variations

Intelligence / Disk Size

Variations

Verified model-weight disk size. Higher intelligence and lower disk size are better.

best67697096.8GB91.8GB86.7GBDisk Size (lower →)Intelligence (higher →)IQ2XXS · antirez/deepseek-v4-gguf · 70% intelligence · 86.7GB disk sizeIQ2XXS · antirez/deepseek-v4-ggufDefault · Reasoning · High · 67% intelligence · 93.7GB disk sizeDefault · Reasoning · HighUD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF · 70% intelligence · 96.8GB disk sizeUD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF
Intelligence Density

Intelligence ÷ Disk Size (GB).

  1. 0.80
    IQ2XXS · antirez/deepseek-v4-gguf
  2. 0.73
    UD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF
  3. 0.71
    Default · Reasoning · High
Hardware

Cost / task speed

Hardware cost / task speed
3 of 3 variations

Model variations

·
6670
Any intelligence
best8m 24s16m 23s31m 57s$6.8k$5.7k$4.7kHardware cost (lower →)Task time (faster →)UD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF (70) · NVIDIA DGX Spark 128GB · llamacpp · $4.7k · 13m 32sSparkUD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF (70) · MacBook Pro M5 Max 128GB · 40-core GPU · llamacpp · $6.7k · 8m 24sM5 Max (40C)UD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF (70) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 31m 57sM3 Ultra (80C)UD-Q2_K_XL · unsloth/DeepSeek-V4-Flash-GGUF (70)IQ2XXS · antirez/deepseek-v4-gguf (70) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 15m 09sM3 Ultra (80C)IQ2XXS · antirez/deepseek-v4-gguf (70) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 11m 33sM5 Max (40C)IQ2XXS · antirez/deepseek-v4-gguf (70)Default · Reasoning · High (67) · NVIDIA DGX Spark 128GB · vllm · $4.7k · 14m 03sSparkDefault · Reasoning · High (67)

Each coloured line is that variation’s cost / task-speed Pareto frontier. Global is the frontier across every qualifying variation; every point is still one original measured configuration.

Frontier →the ranked leaderboardAll models →every measured modelHardware →every measured device
local.aiIndependent local-AI benchmarksLive benchmark data
ContactAccountReferralsAbout