gemma-4-26B-A4B-it
by google
26.5B params · image-text-to-text · 1.2k likes · 12.7M downloads
gemma-4-26B-A4B-it is a 26.5B parameter model. At Q4 quantization it requires 13GB of VRAM. It runs comfortably on GeForce RTX 4090 (49 tok/s), GeForce RTX 5090 (74 tok/s), A100 PCIe 80 GB (90 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Novita | 34 tok/s | ||
| Featherless | |||
| deepinfra | 30 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 49 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 74 tok/s | comfortable |
| M4 Max 128GB | 128GB | 18 tok/s | tight |
| M4 Pro 48GB | 48GB | 9 tok/s | tight |
| M4 Pro 24GB | 24GB | 9 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 90 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 182 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 43 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 36 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 35 tok/s | comfortable |