Google's Gemma 3 successor: unified any-to-any with vision, 256K context, Apache-2.0.
chatgeneralvision · sizes: 12B · Apache-2.0
| machine | best fit | verdict | speed |
|---|---|---|---|
| Apple M1 · 8GB | 12B Q4_K_M | ❌ Won't fit | — |
| Apple M1 Pro · 16GB | 12B Q6_K | 🐢 CPU-only (slow) | 1–2 tok/s |
| Apple M2 · 16GB | 12B Q6_K | 🐢 CPU-only (slow) | 0.6–0.9 tok/s |
| Apple M3 Pro · 18GB | 12B Q8_0 | 🐢 CPU-only (slow) | 0.6–1 tok/s |
| Apple M2 Pro · 32GB | 12B Q8_0 | ✅ Runs (tight) | 7–11 tok/s |
| Apple M4 Pro · 48GB | 12B Q8_0 | ✅ Runs comfortably | 9–15 tok/s |
| Apple M3 Max · 36GB | 12B Q8_0 | ✅ Runs comfortably | 12–19 tok/s |
| Apple M2 Max · 64GB | 12B Q8_0 | ✅ Runs comfortably | 12–19 tok/s |
| Apple M4 Max · 128GB | 12B Q8_0 | ✅ Runs comfortably | 15–25 tok/s |
| Apple M2 Ultra · 192GB | 12B Q8_0 | ✅ Runs comfortably | 15–25 tok/s |
| NVIDIA GeForce RTX 3060 · 12GB VRAM | 12B Q4_K_M | ✅ Runs (tight) | 14–25 tok/s |
| NVIDIA GeForce RTX 4070 · 12GB VRAM | 12B Q4_K_M | ✅ Runs (tight) | 19–30 tok/s |
| NVIDIA GeForce RTX 4090 · 24GB VRAM | 12B Q8_0 | ✅ Runs comfortably | 25–45 tok/s |
| NVIDIA GeForce RTX 5080 · 16GB VRAM | 12B Q6_K | ✅ Runs (tight) | 30–50 tok/s |
| NVIDIA GeForce RTX 5090 · 32GB VRAM | 12B Q8_0 | ✅ Runs comfortably | 50–80 tok/s |
| AMD Radeon RX 7900 XTX · 24GB VRAM | 12B Q8_0 | ✅ Runs comfortably | 25–40 tok/s |
Shopping for a machine? What hardware do I need to run Gemma 4? →
Check your exact rig: open the checker →
2 models share Gemma 4's KV-cache geometry — they size memory the same way, so a rig that fits one tends to fit its same-size kin.
Explore the whole catalog by architecture: open the graph →