Kimi K3
moonshotai/Kimi-K3
Parameters
2800B
Context
1024K
Downloads
1.6M
Likes
10.5K
Architecture
moe_transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 6601.1 GB RAM 9901.6 · Disk 8175.0 | 6903.1 GB RAM 10354.7 · Disk 8201.0 | 7218.9 GB RAM 10828.3 · Disk 8240.0 |
| FP88bit | 3302.3 GB RAM 4953.5 · Disk 4107.0 | 3454.4 GB RAM 5181.7 · Disk 4133.0 | 3620.2 GB RAM 5430.4 · Disk 4172.0 |
| INT88bit | 3302.3 GB RAM 4953.5 · Disk 4107.0 | 3454.4 GB RAM 5181.7 · Disk 4133.0 | 3620.2 GB RAM 5430.4 · Disk 4172.0 |
| AWQ4bit | 1714.8 GB RAM 2572.2 · Disk 2149.3 | 1794.8 GB RAM 2692.1 · Disk 2175.3 | 1888.4 GB RAM 2832.6 · Disk 2214.3 |
| GPTQ4bit | 1735.4 GB RAM 2603.1 · Disk 2174.7 | 1816.3 GB RAM 2724.5 · Disk 2200.7 | 1910.9 GB RAM 2866.3 · Disk 2239.7 |
| GGUF4bit | 1776.7 GB RAM 2665.0 · Disk 2225.6 | 1859.4 GB RAM 2789.1 · Disk 2251.6 | 1955.9 GB RAM 2933.8 · Disk 2290.6 |
GPU Recommendations
⚠️ No single GPU fits — check multi-GPU options
This model needs more VRAM than a single GPU can provide. Cluster options below (may still be short):(1794.76 GB VRAM)
8× H200 SXM
1128 GB · short by 666.76 GB
$20,440/mo
💬 Community — real deployment experience
0 Articles · 0 Benchmarks