Run Gemma 4 12B on M6

Gemma 4 12B runs at 19 tok/s on a 32 GB M6, a community run reported by github.com (LM Studio, QAT). It needs about 9 GB of unified memory. It is Google's 12B dense model, released under the Apache 2.0 licence.

Speed
community19
tok/s
First Token
—
seconds
RAM Needed
9
GB minimum
Engine
LM Studio
of this figure
Want to run Gemma 4 12B faster — or step up to bigger models? Find your Mac →

The figure

This is a community run: a real run, reported by its source and linked below. The index records it as given and never adjusts it.

MetricValue
Tokens per secondcommunity19 tok/s
Provenancecommunity run · github.com
QuantizationQAT
EngineLM Studio
Figure dated2026-10
Minimum RAM9 GB
Parameters12B
Chip bandwidth170 GB/s

Every figure for Gemma 4 12B on the M6

Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (9 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.

SpeedQuantEngineMac RAMContextSourceNote
community19.4 tok/s4-bitMLX32 GB1komlx.ai12-core GPU. oMLX v0.7.2, MTP/DFlash off; gemma-4-12B-it-MLX-4bit; 7.1 GB peak.
community19 tok/sQATLM Studio32 GB—github.com—
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Speed by context depth: Gemma 4 12B on the M6

Measured figures only, each from its source’s own session; the index does not estimate long-context or prefill speed, so a missing cell is —. Each @ column is plain decode with that many tokens already in the prompt, from the same session as the row’s short-context figure: decode slows as the KV cache it reads grows. Prefill is prompt processing, the tokens per second a runtime reads a prompt at; with long agent contexts it sets how long you wait before the first token.

RunDecode (row)@4KPrefillTTFT
MLX · 4-bit · 32 GB Mac · omlx.aicommunity19.4 tok/s
1k
community19 tok/scommunity604 tok/s
at 4K
community2.042 s

Setup: run Gemma 4 12B on a Mac

Ollama — install it (brew install ollama or the app from ollama.com), then pull the model by its tag, checked on ollama.com on 2026-11-14:

ollama run gemma4:12b

MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:

pip install -U mlx-lm
mlx_lm.generate --model <repo-id> --prompt "Hello"

Downloads: Ollama page → · Hugging Face model card →

The 19 tok/s figure above came from LM Studio at QAT; other runtimes and quants run at different speeds.

Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.

Gemma 4 12B on other Apple Silicon chips

ChipSpeedQuantEngineBandwidth
M5 Maxestimated64 tok/sQ4_K_MMLX614 GB/s
M4 Maxestimated57 tok/sQ4_K_MMLX546 GB/s
M4 Proestimated30 tok/sQ4_K_MMLX273 GB/s
M4estimated13 tok/sQ4_K_MMLX120 GB/s

For the estimated figures here, the bandwidth formula’s typical error for dense models is ±9%, against 75 measured runs of 20 models (how it is computed).

About Gemma 4 12B

Google's dense 11.95B Gemma 4 (2 Jul 2026, Apache 2.0): text, image and audio input, 256K context. Model card: MMLU-Pro 77.2, GPQA Diamond 78.8, LiveCodeBench v6 72.0 — close to the 26B-A4B at half the memory. About 7 GB at 4-bit: comfortable on a 16 GB Mac, too tight for 8 GB. Measured at ~19 tok/s on a Mac mini M6 (QAT, community run).

Apache 2.0 licence · 12B · 256K context · LLMCheck capability 30/50 · see its leaderboard rank →

Memory by context length

Estimated, not measured: the index’s minimum for Gemma 4 12B (9 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. 8 full layers (1 kv heads x 512) + 40 sliding layers (8 x 256, window 1024); attention_k_eq_v: K and V counted separately (a runtime that stores the shared tensor once needs half). A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M6 Mac, which runs macOS 26 or later): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).

ContextKV cache (16-bit)RAM neededRAM, q8 KVSmallest tier that fitsWithin the default GPU limit
8K~0.5 GB~9.1 GB~9.0 GB16 GB Macsame
32K~0.9 GB~9.5 GB~9.2 GB16 GB Macsame
64K (coding-agent context)~1.4 GB~10.0 GB~9.5 GB16 GB Macsame
128K~2.5 GB~11.1 GB~10.0 GB16 GB Macsame

System requirements

Macs with the M6

🛒 Get a Mac that runs Gemma 4 12B

Gemma 4 12B needs about 9 GB of unified memory. These Apple Silicon Macs have enough to run it.

As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.

Compare More Models

See how Gemma 4 12B stacks up against other models on your specific Mac hardware.

Open Compare Tool Full Leaderboard