gemma-3-12b-it
by google
12.2B params · image-text-to-text · 761 likes · 2.3M downloads
gemma-3-12b-it is a 12.2B parameter model. At Q4 quantization it requires 6GB of VRAM. It runs comfortably on GeForce RTX 4090 (107 tok/s), GeForce RTX 5090 (161 tok/s), M4 Max 128GB (39 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless | |||
| deepinfra | 40 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 107 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 161 tok/s | comfortable |
| M4 Max 128GB | 128GB | 39 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 19 tok/s | tight |
| M4 Pro 24GB | 24GB | 19 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 197 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 397 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 95 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 78 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 76 tok/s | comfortable |