You can deploy this model and start earning money today!

batiai/
gemma-4-31B-it-GGUF

Name: gemma-4-31B-it-GGUF
Availability: InStock

google/gemma-4-31B-it

quantized

Views

No views yet

Gemma 4 31B-it GGUF — Quantized by BatiAI

BatiFlow

Optimized GGUF quantizations of google/gemma-4-31B-it — Google's largest dense Gemma 4 model with native multilingual + vision capabilities. Built directly from official Google BF16 weights by BatiAI for BatiFlow.

Gemma 4 31B-it is a dense 31 B model (every parameter active per token) — different trade-off from 26B-A4B's MoE design (26 B total / 3.8 B active). Pick 31B for deeper reasoning per query, 26B-A4B for higher throughput.

Quick Start

bash

# 24 GB Mac — Smaller, imatrix-calibrated
ollama pull batiai/gemma4-31b:iq3

# 24 GB+ Mac — Imatrix 4-bit (recommended)
ollama pull batiai/gemma4-31b:iq4

# 32 GB+ Mac — Higher quality
ollama pull batiai/gemma4-31b:q4

# 36 GB+ Mac — Highest on-device quality
ollama pull batiai/gemma4-31b:q6

Available Quantizations

Tag	Quant	Recommended For
`:iq3`	IQ3_M (imatrix)	24 GB Mac, smaller footprint
`:iq4`	IQ4_XS (imatrix)	24 GB+ Mac, recommended
`:q4`	Q4_K_M	32 GB+ Mac, higher quality
`:q6`	Q6_K	36 GB+ Mac, near-lossless

Two modes — text-only by default, multimodal opt-in

Upstream Gemma 4 31B-it is multimodal (text + image + video understanding — but no audio; for audio see the smaller E2B / E4B variants). In the GGUF ecosystem this is delivered as two files: a main model.gguf (text tower) and a separate mmproj.gguf (multi-modal projector — the vision tower). We ship both, separate, so you can pick:

	Text-only (default)	Multimodal (opt-in)
Files needed	main GGUF only	main GGUF + `mmproj-*.gguf`
Capabilities	Q&A, coding, tool calling, RAG, agents	+ image / video understanding (OCR, captioning, visual reasoning)
`ollama pull`	✅ single command	⚠ Ollama mmproj integration is still rough — use llama.cpp directly
Disk / RAM	smaller (no vision weights)	larger (+ ~770 MB to ~1.2 GB)
Recommended for	most users (chat, code, agents)	OCR, image understanding

Multimodal usage (llama.cpp)

Download the main GGUF + the mmproj file:

bash

# Pick a main model (text tower)
wget https://huggingface.co/batiai/Gemma-4-31B-it-GGUF/resolve/main/google-gemma-4-31B-it-IQ4_XS.gguf

# Pick the mmproj (vision tower) — Q6_K is the sweet spot, BF16 if you want zero loss
wget https://huggingface.co/batiai/Gemma-4-31B-it-GGUF/resolve/main/mmproj-Q6_K.gguf

Server mode (OpenAI-compatible Vision API):

bash

llama-server \
  -m google-gemma-4-31B-it-IQ4_XS.gguf \
  --mmproj mmproj-Q6_K.gguf \
  -c 32768 --host 127.0.0.1 --port 8080

curl http://127.0.0.1:8080/v1/chat/completions -d '{
  "messages": [{
    "role": "user",
    "content": [
      {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,..."}},
      {"type": "text", "text": "What does this screenshot show?"}
    ]
  }]
}'

One-shot CLI:

bash

llama-mtmd-cli \
  -m google-gemma-4-31B-it-IQ4_XS.gguf \
  --mmproj mmproj-Q6_K.gguf \
  --image ~/Desktop/photo.jpg \
  -p "describe this image"

mmproj quantizations available

File	Quant	Size	When to use
`mmproj-Q6_K.gguf`	Q6_K	~770 MB	balanced (recommended)
`mmproj-BF16.gguf`	BF16	~1.2 GB	absolute zero quantization loss for vision

(Q8_0 not available because some vision tensors have shapes incompatible with Q8_0's column alignment — applies to every quantizer of this model. Q6_K's K-quant block layout handles them.)

Dense vs MoE — when to pick 31B vs 26B-A4B?

	31B-it (dense)	26B-A4B-it (MoE)
Active params / token	31 B (all)	3.8 B (8 experts of 32)
Throughput	slower (more compute per token)	faster
Reasoning depth	better (denser computation)	good
Memory at IQ4	~17 GB	~13 GB
Best for	hard reasoning / single-thread workloads	high-throughput / agents / streaming chat

Use 31B for quality-per-token, 26B-A4B for tokens-per-second.

Why BatiAI?

	BatiAI	Third-party (unsloth, etc.)
Source	Quantized directly from official Google weights	Often re-quantized from other GGUFs
Compatibility	✅ Verified on Ollama 0.20+	❌ Known issues with Ollama 0.20+
Tested on	Real Mac mini M4 + MacBook Pro M4 Max	Often untested
Tool calling	✅ Verified with BatiFlow's 57 tool functions	Often broken
Korean	✅ Validated	Not tested
Multimodal	✅ Vision mmproj available (this repo)	Often missing
Signing	`general.author: BatiAI` for provenance	Varies

About BatiFlow

flow.bati.ai

BatiFlow is a macOS-native AI desktop automation app — just 5 MB, built with Swift.

Free & Unlimited — On-device AI via Ollama, no API costs
100 % Private — All data stays on your Mac
Ultra Lightweight — Native macOS app, only 5 MB
57 built-in tools — calendar, notes, reminders, files, email, browser, messaging, and more

Related models in the BatiAI Gemma 4 lineup

Model	Modalities	Min RAM	Repo
Gemma 4 E2B-it	text + image + audio	8 GB	batiai/Gemma-4-E2B-it-GGUF
Gemma 4 E4B-it	text + image + audio	8 GB	batiai/Gemma-4-E4B-it-GGUF
Gemma 4 26B-A4B-it	text + image / video (MoE)	24 GB	batiai/Gemma-4-26B-A4B-it-GGUF
Gemma 4 31B-it	text + image / video (dense)	24 GB	this repo

Technical Details

Original Model: google/gemma-4-31B-it
Architecture: Gemma 4 dense (31 B parameters, all active per token)
Modalities: Text (primary) + Image / Video via opt-in mmproj. Audio not supported (use E2B / E4B for audio)
Context Window: 128 K tokens
License: Gemma
Quantized with: llama.cpp
Quantized by: BatiAI
GGUF metadata: general.author: BatiAI, general.url: https://flow.bati.ai

License

Mirrors the upstream Gemma license. Commercial use permitted per Google's Gemma terms.

BatiAI's quantization pipeline is provided under MIT.

batiai/gemma-4-31B-it-GGUF

Gemma 4 31B-it GGUF — Quantized by BatiAI

Quick Start

Available Quantizations

Two modes — text-only by default, multimodal opt-in

Multimodal usage (llama.cpp)

mmproj quantizations available

Dense vs MoE — when to pick 31B vs 26B-A4B?

Why BatiAI?

About BatiFlow

Related models in the BatiAI Gemma 4 lineup

Technical Details

License

batiai/
gemma-4-31B-it-GGUF