Gemma 4 — local AI model by Google
Written by Jakub Rusinowski · Last updated
Google's fourth-generation open model family. The first wave (E2B, E4B, 26B-A4B MoE, 31B dense) launched March 31, 2026 under Apache 2.0 with a fully multimodal architecture — text, image, and audio across all sizes, plus video on the 31B. A separate 12B 'Unified' encoder-free model followed June 3, 2026, adding native audio+video to a laptop-friendly 16GB footprint. The E2B and E4B Efficient variants run on 4–6 GB VRAM GPUs, while the 26B-A4B MoE and 31B dense models deliver near-frontier quality on a single RTX 4090.
Variants
The smallest Gemma 4 variant needs about 4 GB of VRAM at Q4 (QAT) — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Gemma 4 E2B → 2B (Efficient) | ~3.9 GB | 128,000 | ollama run gemma4:e2b | |
| Gemma 4 E4B → 4B (Efficient) | ~5.6 GB | 128,000 | ollama run gemma4:e4b | |
| Gemma 4 26B-A4B → 26B (4B active, MoE) | ~16.5 GB | 256,000 | ollama run gemma4:26b-a4b | |
| Gemma 4 31B → 31B | ~19.5 GB | 256,000 | ollama run gemma4:31b | |
| Gemma 4 12B (Unified) → 12B | ~8 GB | 128,000 | ollama run gemma4:12b |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Gemma 4 locally
Install Ollama, then pull the tag.
ollama run gemma4:e2bPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Gemma 4 E2B, Gemma 4 E4B, Gemma 4 26B-A4B, Gemma 4 31B, Gemma 4 12B (Unified)Recommended GPU
The cheapest catalogued GPU that runs Gemma 4 locally (min 4 GB VRAM) is the Intel Arc B570 (10 GB).
Can I run Gemma 4 on my GPU?
- Gemma 4 on AMD Radeon RX 7800 XT
- Gemma 4 on AMD Radeon RX 7900 XT
- Gemma 4 on AMD Radeon RX 7900 XTX
- Gemma 4 on AMD Radeon RX 9060 XT 16GB
- Gemma 4 on AMD Radeon RX 9060 XT 8GB
- Gemma 4 on AMD Radeon RX 9070
- Gemma 4 on AMD Radeon RX 9070 XT
- Gemma 4 on Apple M1
- Gemma 4 on Apple M1 Pro
- Gemma 4 on Apple M2
- Gemma 4 on Apple M2 Pro
- Gemma 4 on Apple M3
- Gemma 4 on Apple M3 Pro
- Gemma 4 on Apple M4
- Gemma 4 on Apple M4 Pro
- Gemma 4 on Apple M5
- Gemma 4 on Intel Arc B570
- Gemma 4 on Intel Arc B580
- Gemma 4 on NVIDIA GeForce RTX 3060 (12GB)
- Gemma 4 on NVIDIA GeForce RTX 3070
- Gemma 4 on NVIDIA GeForce RTX 3070 Ti
- Gemma 4 on NVIDIA GeForce RTX 3080 (10GB)
- Gemma 4 on NVIDIA GeForce RTX 3090
- Gemma 4 on NVIDIA GeForce RTX 4060
- Gemma 4 on NVIDIA GeForce RTX 4060 Ti 16GB
- Gemma 4 on NVIDIA GeForce RTX 4070
- Gemma 4 on NVIDIA GeForce RTX 4070 Super
- Gemma 4 on NVIDIA GeForce RTX 4070 Ti
- Gemma 4 on NVIDIA GeForce RTX 4070 Ti Super
- Gemma 4 on NVIDIA GeForce RTX 4080
- Gemma 4 on NVIDIA GeForce RTX 4080 Super
- Gemma 4 on NVIDIA GeForce RTX 4090
- Gemma 4 on NVIDIA GeForce RTX 5060
- Gemma 4 on NVIDIA GeForce RTX 5060 Ti 16GB
- Gemma 4 on NVIDIA GeForce RTX 5060 Ti 8GB
- Gemma 4 on NVIDIA GeForce RTX 5070
- Gemma 4 on NVIDIA GeForce RTX 5070 Ti
- Gemma 4 on NVIDIA GeForce RTX 5080
- Gemma 4 on NVIDIA GeForce RTX 5090
Gemma 4 — frequently asked questions
How much VRAM does Gemma 4 need?
Gemma 4 needs about 4 GB VRAM at Q4 (QAT) quantization for its smallest variant. Variants: Gemma 4 E2B (4 GB, Q4 (QAT)); Gemma 4 E4B (6 GB, Q4 (QAT)); Gemma 4 26B-A4B (16 GB, Q4_K_M); Gemma 4 31B (20 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Gemma 4 on an RTX 4090 (24 GB)?
Yes — Gemma 4 runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4 (QAT).
What quantization should I use for Gemma 4?
Q4_K_M is the best balance of quality and VRAM for Gemma 4 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Gemma 4 with Ollama?
Install Ollama, then run: ollama run gemma4:e2b. This downloads Gemma 4 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.