command-a-plus-05-2026-fp8
by CohereLabs
219B params · image-text-to-text · 36 likes · 7.0k downloads
command-a-plus-05-2026-fp8 is a 219B parameter model. At Q4 quantization it requires 109GB of VRAM. It runs comfortably on H200 SXM 141 GB (32 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Cohere | 165 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| H200 SXM 141 GB | 141GB | 32 tok/s | comfortable |
| M2 Ultra 192GB | 192GB | 3 tok/s | tight |