Gemma4All logoGemma4All

Can You Run Gemma 4?

Pick your device below and get an instant verdict for every Gemma 4 model, from the tiny E2B up to the 31B flagship. The numbers come straight from Google's official documentation, not guesswork.

Estimated usable memory for model weights: 11.0 GB (total minus ~5 GB reserved for macOS)

Gemma 4 E2B

Comfortable · BF16 (16-bit)

Smooth — plenty of headroom for long conversations and multitasking.

Weights: ~8 GB

Gemma 4 E4B

Comfortable · SFP8 (8-bit)

Smooth — plenty of headroom for long conversations and multitasking.

Weights: ~7.5 GB

Recommended

Gemma 4 12B

Comfortable · Q4_0 (4-bit)

Smooth — plenty of headroom for long conversations and multitasking.

Weights: ~6.7 GB

Only mid-size model with native audio input — speech recognition and diarization built in.

Grab the official QAT build (~7.2 GB) for noticeably better quality at essentially the same size. See our QAT guide.

Gemma 4 26B A4B

Won't fit

Won't fit on this device, even at the lightest quant.

Gemma 4 31B

Won't fit

Won't fit on this device, even at the lightest quant.

How this works

Every number in this checker comes from Google's official Gemma 4 documentation, via our hardware requirements guide. The checker walks each model's quantization levels from highest quality to lowest — BF16, then SFP8, then Q4_0 — and reports the best one that fits your usable memory:

  • Comfortable — weights use no more than 80% of your usable memory, leaving headroom for the KV cache and multitasking.
  • Tight — weights fit within your usable memory, but with little room to spare.
  • Won't fit — even the smallest quant is too large for your device.

Estimating "usable memory" is deliberately conservative, per platform:

  • Mac (unified memory): total memory minus ~5 GB reserved for macOS and background processes.
  • NVIDIA GPU: total VRAM minus ~1 GB reserved by the driver and OS.
  • CPU-only: total system RAM ÷ 2 — one copy of the weights, plus working memory for computation.

These are the same rules of thumb used throughout our hardware guide, so the results here match what you'll read in the rest of the site.

FAQ

Can my 16GB MacBook run Gemma 4?

Yes. A 16 GB Mac has about 11 GB of usable memory after macOS overhead, which comfortably fits Gemma 4 12B at Q4_0 (~6.7 GB) with room for a long conversation's KV cache. E4B and E2B run comfortably too. The 26B A4B model is too tight at 16 GB — it needs 24 GB or more.

What's the minimum to run Gemma 4 at all?

8 GB total memory is the practical floor. On an 8 GB Mac that leaves only 3 GB usable, which is very tight even for the smallest E2B model at Q4_0 (~2.5 GB). 16 GB is a much more comfortable starting point for anything beyond quick tests.

Do I need a GPU?

No. Gemma 4 runs on CPU alone using system RAM, following the RAM ÷ 2 rule (one copy for weights, one for working memory) — but inference is much slower, often just a few tokens per second. The 26B A4B model is a notable exception: only ~4B of its parameters activate per token, making it a surprisingly capable CPU-only pick if you have 32 GB or more.

What about the QAT versions?

Google shipped official quantization-aware training (QAT) checkpoints for E2B, E4B, 12B, and 31B. They land at essentially the same download size as regular Q4_0 but hold quality much better, so grab the QAT build whenever Q4_0 is your best-fitting quant. The 26B A4B model does have an official Q4_0 QAT checkpoint, but its narrow expert layers lose more accuracy at 4-bit — a mixed-precision community quant is the better call there. See our dedicated QAT guide for exact file names.

Can Gemma 4 run on my phone?

Yes, via the official Google AI Edge Gallery app (LiteRT mobile builds) — but only the E2B and E4B models are built for phones. E2B needs about 6 GB of RAM to run (comfortable at 8 GB+), and E4B needs about 8 GB (comfortable at 12 GB+). That covers most recent iPhones and Android flagships. See our full phone guide for setup steps and device-by-device notes.

Related guides