local.aiEarly access
HardwareModelsEnergyWallReferralsContact

Search

Find people, models, hardware, or pages

Account
HardwareModelsEnergyWallReferralsContactClaim your nameAccount
Model

Step 3.7 Flash

Intelligence rank
#22/238
70
Speed rank
#53/238
1m 29s
Fastest hardware: NVIDIA RTX PRO 6000 Blackwell 96GB.

Quick Start

Variations

Intelligence / Disk Size

Variations

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

best596570118GB90.0GB61.7GBDisk Size (lower →)Intelligence (higher →)UD-IQ2_XXS · 67% intelligence · 61.7GB disk sizeUD-IQ2_XXSUD-IQ2_M · unsloth/Step-3.7-Flash-GGUF · 59% intelligence · 61.8GB disk sizeUD-IQ2_M · unsloth/Step-3.7-Flash-GGUFIQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF · 59% intelligence · 75.8GB disk sizeIQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUFQ3_K_M · 69% intelligence · 93.8GB disk sizeQ3_K_MDefault · 0xSero/Step-3.7-Flash-148B · 63% intelligence · 101GB disk sizeDefault · 0xSero/Step-3.7-Flash-148BQ3_K_L · 64% intelligence · 103GB disk sizeQ3_K_LIQ4_XS · 70% intelligence · 105GB disk sizeIQ4_XSQ4_K_S · 69% intelligence · 111GB disk sizeQ4_K_SMXFP4_MOE · 64% intelligence · 118GB disk sizeMXFP4_MOE
Intelligence Density

Intelligence ÷ Disk Size (GB).

  1. 1.09
    UD-IQ2_XXS
  2. 0.96
    UD-IQ2_M · unsloth/Step-3.7-Flash-GGUF
  3. 0.78
    IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF
  4. 0.74
    Q3_K_M
  5. 0.67
    IQ4_XS
  6. 0.62
    Default · 0xSero/Step-3.7-Flash-148B
  7. 0.62
    Q4_K_S
  8. 0.62
    Q3_K_L
  9. 0.54
    MXFP4_MOE
Hardware

Cost / task speed

Hardware cost / task speed
9 of 9 variations

Model variations

·
5970
Any intelligence
best1m 29s4m 40s14m 45s$16.3k$8.7k$4.7kHardware cost (lower →)Task time (faster →)UD-IQ2_XXS (67) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 8m 41sSparkUD-IQ2_XXS (67) · Mac Studio M3 Ultra 96GB · 80-core GPU · llama_cpp · $6.8k · 5mM3 Ultra (80C)UD-IQ2_XXS (67) · Mac Studio M3 Ultra 96GB · 60-core GPU · llama_cpp · $5.3k · 5m 55sM3 Ultra (60C)UD-IQ2_XXS (67) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 4m 43sM5 Max (40C)UD-IQ2_XXS (67) · NVIDIA RTX PRO 6000 Blackwell 96GB · llamacpp · $16.3k · 1m 29sRTXPro6kUD-IQ2_XXS (67)IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 8m 46sSparkIQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · Mac Studio M3 Ultra 96GB · 80-core GPU · llama_cpp · $6.8k · 5m 04sM3 Ultra (80C)IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · Mac Studio M3 Ultra 96GB · 60-core GPU · llama_cpp · $5.3k · 6m 07sM3 Ultra (60C)IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · MacBook Pro M4 Max 128GB · 40-core GPU · llama_cpp · $5.6k · 6m 55sM4 Max (40C)IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 4m 54sM5 Max (40C)IQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59) · NVIDIA RTX PRO 6000 Blackwell 96GB · llama_cpp · $16.3k · 1m 32sRTXPro6kIQ3_XXS · stepfun-ai/Step-3.7-Flash-GGUF (59)UD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 10m 50sSparkUD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59) · Mac Studio M3 Ultra 96GB · 80-core GPU · llama_cpp · $6.8k · 5m 31sM3 Ultra (80C)UD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59) · Mac Studio M3 Ultra 96GB · 60-core GPU · llama_cpp · $5.3k · 6m 36sM3 Ultra (60C)UD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 5m 23sM5 Max (40C)UD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59) · NVIDIA RTX PRO 6000 Blackwell 96GB · llamacpp · $16.3k · 1m 48sRTXPro6kUD-IQ2_M · unsloth/Step-3.7-Flash-GGUF (59)IQ4_XS (70) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 8m 08sSparkIQ4_XS (70) · Mac Studio M3 Ultra 96GB · 80-core GPU · llama_cpp · $6.8k · 4m 22sM3 Ultra (80C)IQ4_XS (70) · MacBook Pro M4 Max 128GB · 40-core GPU · llama_cpp · $5.6k · 5m 57sM4 Max (40C)IQ4_XS (70) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 4m 24sM5 Max (40C)IQ4_XS (70)Q4_K_S (69) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 8m 50sSparkQ4_K_S (69) · MacBook Pro M4 Max 128GB · 40-core GPU · llama_cpp · $5.6k · 5m 59sM4 Max (40C)Q4_K_S (69) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 4m 25sM5 Max (40C)Q4_K_S (69) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 4m 39sM3 Ultra (80C)Q4_K_S (69)Q3_K_L (64) · NVIDIA DGX Spark 128GB · llama_cpp · $4.7k · 10m 09sSparkQ3_K_L (64) · MacBook Pro M5 Max 128GB · 40-core GPU · llama_cpp · $6.7k · 5m 17sM5 Max (40C)Q3_K_L (64) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 5m 44sM3 Ultra (80C)Q3_K_L (64)Q3_K_M (69) · NVIDIA DGX Spark 128GB · llamacpp · $4.7k · 10m 32sSparkQ3_K_M (69) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 5m 39sM3 Ultra (80C)Q3_K_M (69) · MacBook Pro M4 Max 128GB · 40-core GPU · llamacpp · $5.6k · 8m 02sM4 Max (40C)Q3_K_M (69) · MacBook Pro M5 Max 128GB · 40-core GPU · llamacpp · $6.7k · 5m 20sM5 Max (40C)Q3_K_M (69)MXFP4_MOE (64) · NVIDIA DGX Spark 128GB · llamacpp · $4.7k · 12m 45sSparkMXFP4_MOE (64) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 5m 21sM3 Ultra (80C)MXFP4_MOE (64)Default · 0xSero/Step-3.7-Flash-148B (63) · NVIDIA DGX Spark 128GB · vllm · $4.7k · 14m 45sSparkDefault · 0xSero/Step-3.7-Flash-148B (63)

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