Google · 26B (4B active) · Mixture of Experts
Discrete diffusion MoE — 1100+ tok/s on H100, multimodal (text/image/video)
1.9M downloads
1.2K likes
2026-06 256K context
Use Cases
chat vision reasoning
Mixture of Experts
Total experts: 128
Active experts: 8
Active params: 3.8B
| Quant | Bits | VRAM | Quality | Status |
|---|---|---|---|---|
| Q2_K | 2 | 8.8 GB | low | — |
| Q3_K_M | 3 | 12.2 GB | moderate | — |
| Q4_K_M | 4 | 13.8 GB | good | — |
| Q5_K_M | 5 | 17.1 GB | good | — |
| Q6_K | 6 | 20.5 GB | excellent | — |
| Q8_0 | 8 | 27.1 GB | excellent | — |
| F16 | 16 | 53.8 GB | lossless | — |
GPUs that run DiffusionGemma 26B-A4B IT
DiffusionGemma 26B-A4B IT needs about 13.8 GB of VRAM at Q4_K_M, so it runs comfortably on a GPU with 20 GB of VRAM or more (39 of the GPUs we track). Common cards, smallest first:
| VRAM | GPUs |
|---|---|
| 20 GB | RTX 4000 SFF Ada , RX 7900 XT , RTX 4000 Ada |
| 24 GB | RTX 4090 , RX 7900 XTX , RTX 3090 Ti |
| 32 GB | RTX 5090 , RTX 5000 Ada , Tesla V100 32GB |
About this model
Discrete diffusion MoE — 1100+ tok/s on H100, multimodal (text/image/video)