Kimi K2.5
moonshotai/Kimi-K2.5
Parameters
1040B
Context
256K
Downloads
856K
Likes
2.9K
Architecture
moe_transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 2455.5 GB RAM 3683.2 · Disk 3061.0 | 2582.6 GB RAM 3873.9 · Disk 3087.0 | 2790.9 GB RAM 4186.3 · Disk 3126.0 |
| FP88bit | 1230.2 GB RAM 1845.3 · Disk 1550.0 | 1301.6 GB RAM 1952.5 · Disk 1576.0 | 1454.2 GB RAM 2181.4 · Disk 1615.0 |
| INT88bit | 1230.2 GB RAM 1845.3 · Disk 1550.0 | 1301.6 GB RAM 1952.5 · Disk 1576.0 | 1454.2 GB RAM 2181.4 · Disk 1615.0 |
| AWQ4bit | 640.6 GB RAM 960.8 · Disk 822.8 | 685.2 GB RAM 1027.8 · Disk 848.8 | 811.0 GB RAM 1216.5 · Disk 887.8 |
| GPTQ4bit | 648.2 GB RAM 972.3 · Disk 832.3 | 693.2 GB RAM 1039.8 · Disk 858.3 | 819.3 GB RAM 1229.0 · Disk 897.3 |
| GGUF4bit | 663.5 GB RAM 995.3 · Disk 851.2 | 709.2 GB RAM 1063.8 · Disk 877.2 | 836.0 GB RAM 1254.1 · Disk 916.2 |
GPU Recommendations
Best ValueBEST
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 30.6 tok/s
Effective production throughput ≈ 21.4 tok/s
score 0.3817
Cheapest
4× B200
720 GB VRAM
$12,051/mo
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.2967
Performance
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 30.6 tok/s
Effective production throughput ≈ 21.4 tok/s
score 0.3817
Min Complexity
4× B200
720 GB VRAM
$12,051/mo
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.2967
💬 Community — real deployment experience
0 Articles · 0 Benchmarks