back to local model checker

Gemma 4 31B

Vision

Google · 33B · Dense

Gemma 4 flagship base model (official)

Gemma
Downloads

0

Likes

0

Released

2026-04

Context

256K tokens

Use cases

visionreasoning

Your hardware

16 GB Reference Desktop

Quant
Q4_K_M
Memory
17.4 GB
Speed
~7 tok/s
Fit
Too heavy

Quantization options

What is quantization?
QuantBitsVRAMDiskQualityStatus
Q2_K211.1 GB10.1 GBlowTight fit · ~18 t/s
Q3_K_M315.3 GB14.1 GBmoderateBarely runs · ~13 t/s
Q4_K_M417.4 GB16.1 GBgoodToo heavy · ~7 t/s
Q5_K_M521.6 GB20.2 GBgoodBarely runs · ~4 t/s
Q6_K625.9 GB24.2 GBexcellentBarely runs · ~2 t/s
Q8_0834.3 GB32.3 GBexcellentBarely runs · ~1 t/s
F161668.1 GB64.5 GBlosslessToo heavy · ~1 t/s

About this model

Gemma 4 31B (Gemma 4 31b) is a 33B dense transformer from Google in the Gemma family, designed primarily for vision, reasoning workflows. Supports up to 256K context tokens; start at 4K–8K with Q8_0 KV cache on tight GPUs.

Hardware planning baselines start at 18.4 GB minimum memory and 30.7 GB recommended memory. At Q4_K_M, weights plus baseline runtime overhead use roughly 17.4 GB of RAM/VRAM and 16.1 GB of disk space. As a 33B dense checkpoint, every parameter participates in each token forward pass, meaning decoding speed scales linearly with GPU memory bandwidth and drops sharply if layers spill over PCIe into host DDR5 RAM.

Deploy Gemma 4 31B locally via llama.cpp with Flash Attention enabled. Under Gemma Terms of Use, review official weight cards before commercial redistribution.

Highlights

  • Local Fit Baseline18.4 GB minimum and 30.7 GB recommended RAM/VRAM.
  • Primary QuantQ4_K_M targets 17.4 GB memory and 16.1 GB disk.
  • Context & Capabilities256K max context · Dense.

System requirements by device

Memory planning targets 18.4 GB minimum RAM/VRAM and 30.7 GB recommended memory. The baseline quant is Q4_K_M, estimated at 17.4 GB memory and 16.1 GB disk. Actual runtime usage shifts with context length, KV cache precision, and OS background overhead.

DeviceNotes
Mac (Apple silicon)Use unified memory headroom (~75% physical RAM max) as the primary fit signal. Start with Q4_K_M and close heavy browser tabs first.
Windows / Linux GPUKeep weights plus KV cache inside dedicated VRAM so Windows WDDM does not spill layers over PCIe into Shared GPU Memory.
Phones & Compact Edge BoardsNot practical on phones (18.4 GB min RAM exceeds Android LMK and iOS Jetsam limits). Pick a 2B–4B model on mobile.

Start with a short session and leave memory for the operating system. A file fitting on disk does not guarantee that it will load in memory.

Estimated memory by quantization

These are rough estimates, not tested minimum requirements. Check the actual file and runtime documentation; context, offloading and operating-system overhead can change the memory needed.

Quant TierEstimated memoryDisk SizeQuality
Q2_K11.1 GB10.1 GBlow
Q3_K_M15.3 GB14.1 GBmoderate
Q4_K_M17.4 GB16.1 GBgood
Q5_K_M21.6 GB20.2 GBgood
Q6_K25.9 GB24.2 GBexcellent
Q8_034.3 GB32.3 GBexcellent
F1668.1 GB64.5 GBlossless

How to run Gemma 4 31B on your device

Start with a short prompt and the smallest quant that still gives the quality you need. A clean first run is more useful than loading a larger file with no memory headroom.

Mac

Run Gemma 4 31B on Mac

  1. Install LM Studio or llama.cpp on macOS Apple Silicon.
  2. Pull Q4_K_M (17.4 GB) first and confirm it stays below ~75% of your Mac's unified memory.
  3. Enable Flash Attention and cap initial context at 8K until memory pressure stays below 30.7 GB.
  4. Consult the Mac guide for Activity Monitor tuning and Metal offload tips.
Device how-to guide
Windows

Run Gemma 4 31B on Windows

  1. Install LM Studio for Windows for visual VRAM inspection, or Ollama when you want a localhost API endpoint.
  2. Load Q4_K_M (~17.4 GB) and verify Dedicated GPU Memory stays within VRAM without spilling into Shared GPU Memory.
  3. Reserve at least 18.4 GB memory minimum and target 30.7 GB for comfortable context growth.
  4. Enable Q8_0 KV cache or drop one quant tier if generation slows down over PCIe.
Device how-to guide

Gemma 4 31B vs alternatives

ModelParamsContextMin RAMLicense
Gemma 4 31B33B256K18.4 GBGemma
Gemma 4 31B IT33B256K18.4 GBGemma
Gemma 2 27B27B8K15.1 GBGemma
Gemma 3 27B27B128K15.1 GBGemma
Gemma 4 26B-A4B IT27B (4B active)256K15.1 GBGemma
Gemma 4 26B-A4B27B (4B active)256K15.1 GBGemma

Use this table to choose the smallest nearby model that still matches the job. A related model with lower RAM can be the better local choice when the current page fits only at the edge of available memory.

Frequently asked questions

What is Gemma 4 31B?
Gemma 4 31B is a 33B dense transformer from Google, supporting up to 256K tokens with 18.4 GB minimum RAM and 30.7 GB recommended RAM.
What is the best way to run Gemma 4 31b locally?
Choose the smallest quant that matches the task, test one short prompt, then compare RAM, context, and nearby models before making Gemma 4 31B your default local model.
How much VRAM or RAM does Gemma 4 31B require?
At Q4_K_M, Gemma 4 31B occupies roughly 17.4 GB of active memory and 16.1 GB on disk. Add 2–6 GB of spare headroom for KV cache buffers, runtime scratch space, and OS apps (30.7 GB recommended).
What is the best way to run Gemma 4 31B locally?
Run Gemma 4 31B in llama.cpp starting with Q4_K_M. Keep initial context at 4K–8K tokens and confirm 100% GPU/Metal offload before scaling context length.
Can Gemma 4 31B run on Android or iPhone?
Gemma 4 31B requires 18.4 GB minimum memory, placing it outside practical 8 GB smartphone limits. Run it on a 16 GB+ Mac or desktop GPU instead, and pick a 2B–4B checkpoint for phones.
How does Gemma 4 31B compare to Gemma 4 31B IT?
Gemma 4 31B IT (33B, 18.4 GB min RAM) sits closest in size. Compare active parameters, context ceiling (256K vs 256K), and license terms (Gemma) before switching.