local.aiEarly access
HardwareModelsEnergyWallReferralsContact

Search

Find people, models, hardware, or pages

Account
HardwareModelsEnergyWallReferralsContactClaim your nameAccount
Model

Gemma 4 E4B It

Intelligence rank
#102/238
29
Speed rank
#6/238
12s
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.

best25272913.3GB9.1GB5.0GBDisk Size (lower →)Intelligence (higher →)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF · 29% intelligence · 5.0GB disk sizeQ4_K_M · unsloth/gemma-4-E4B-it-GGUFfp8 · prithivMLmods/gemma-4-E4B-it-FP8 · 27% intelligence · 13.3GB disk sizefp8 · prithivMLmods/gemma-4-E4B-it-FP8fp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic · 25% intelligence · 13.3GB disk sizefp8 · leon-se/gemma-4-E4B-it-FP8-Dynamicfp8 · Draft Model · k=4 · 25% intelligence · 13.3GB disk sizefp8 · Draft Model · k=4
Intelligence Density

Intelligence ÷ Disk Size (GB).

  1. 5.80
    Q4_K_M · unsloth/gemma-4-E4B-it-GGUF
  2. 2.00
    fp8 · prithivMLmods/gemma-4-E4B-it-FP8
  3. 1.87
    fp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic
  4. 1.87
    fp8 · Draft Model · k=4
Hardware

Cost / task speed

Hardware cost / task speed
4 of 4 variations

Model variations

·
2428
Any intelligence
best12s33s1m 33s$15.5k$6.5k$2.7kHardware cost (lower →)Task time (faster →)fp8 · Draft Model · k=4 (25) · NVIDIA DGX Spark 128GB · vllm · $4.7k · 45sSparkfp8 · Draft Model · k=4 (25) · NVIDIA GeForce RTX 4090 24GB · vllm · $3.4k · 17sRTX 4090fp8 · Draft Model · k=4 (25) · NVIDIA GeForce RTX 5090 32GB · vllm · $4.8k · 12sRTX 5090fp8 · Draft Model · k=4 (25) · NVIDIA RTX 6000 Ada 48GB · vllm · $8.2k · 16sRTX6k Adafp8 · Draft Model · k=4 (25) · NVIDIA RTX PRO 6000 Blackwell 96GB · vllm · $15.5k · 12sRTXPro6kfp8 · Draft Model · k=4 (25)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · NVIDIA DGX Spark 128GB · llamacpp · $4.7k · 56sSparkQ4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · Mac Studio M3 Ultra 96GB · 80-core GPU · llamacpp · $6.8k · 41sM3 Ultra (80C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · Mac Studio M3 Ultra 96GB · 60-core GPU · llamacpp · $5.3k · 43sM3 Ultra (60C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · MacBook Pro M4 Max 36GB · 32-core GPU · llamacpp · $3.2k · 52sM4 Max (32C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · Mac mini M4 Pro 48GB · 20-core GPU · llamacpp · $2.7k · 1m 15sM4 Pro (20C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · MacBook Pro M5 Max 128GB · 40-core GPU · llamacpp · $6.7k · 33sM5 Max (40C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · MacBook Pro M5 Pro 64GB · 20-core GPU · llamacpp · $3.7k · 55sM5 Pro (20C)Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · NVIDIA GeForce RTX 4090 24GB · llamacpp · $3.4k · 20sRTX 4090Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · NVIDIA GeForce RTX 5090 32GB · llamacpp · $4.8k · 14sRTX 5090Q4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · NVIDIA RTX 6000 Ada 48GB · llamacpp · $8.2k · 19sRTX6k AdaQ4_K_M · unsloth/gemma-4-E4B-it-GGUF (29) · NVIDIA RTX PRO 6000 Blackwell 96GB · llamacpp · $15.5k · 15sRTXPro6kQ4_K_M · unsloth/gemma-4-E4B-it-GGUF (29)fp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25) · NVIDIA DGX Spark 128GB · vllm · $4.7k · 1m 32sSparkfp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25) · NVIDIA GeForce RTX 4090 24GB · vllm · $3.4k · 30sRTX 4090fp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25) · NVIDIA GeForce RTX 5090 32GB · vllm · $4.8k · 18sRTX 5090fp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25) · NVIDIA RTX 6000 Ada 48GB · vllm · $8.2k · 31sRTX6k Adafp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25) · NVIDIA RTX PRO 6000 Blackwell 96GB · vllm · $15.5k · 18sRTXPro6kfp8 · leon-se/gemma-4-E4B-it-FP8-Dynamic (25)fp8 · prithivMLmods/gemma-4-E4B-it-FP8 (27) · NVIDIA DGX Spark 128GB · vllm · $4.7k · 1m 33sSparkfp8 · prithivMLmods/gemma-4-E4B-it-FP8 (27) · NVIDIA GeForce RTX 5090 32GB · vllm · $4.8k · 32sRTX 5090fp8 · prithivMLmods/gemma-4-E4B-it-FP8 (27) · NVIDIA RTX 6000 Ada 48GB · vllm · $8.2k · 32sRTX6k Adafp8 · prithivMLmods/gemma-4-E4B-it-FP8 (27) · NVIDIA RTX PRO 6000 Blackwell 96GB · vllm · $15.5k · 19sRTXPro6kfp8 · prithivMLmods/gemma-4-E4B-it-FP8 (27)

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