How much VRAM does Kimi K3 need?
Kimi K3 needs 1,454 GB for its weights as released (INT4), 1,455 GB with an 8K context and 1,456 GB with 32K.
- Parameters
- 2.8T
- Active per token
- 104B
- Longest context
- 1M
- Released
- Jun 2026
- License
- Kimi K3 License
- Strongest at
- Chat, Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (INT4) | 1,454 GB | 1,455 GB | 1,456 GB | 1,459 GB |
| 3-bit | 1,265 GB | 1,267 GB | 1,268 GB | 1,270 GB |
| 2-bit | 971 GB | 973 GB | 973 GB | 976 GB |
| 1.6-bit | 534 GB | 536 GB | 536 GB | 539 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | Too big | |||
| RTX PRO 6000 Blackwell | Too big | |||
| M4 Max 128 GB | Too big | |||
| M5 Max 128 GB | Too big | |||
| Ryzen AI Max+ 395 128 GB | Too big | |||
| DGX Spark 128 GB | Too big | |||
| M3 Ultra 256 GB | Too big | |||
| M5 Ultra 512 GB | Too big | |||
| H100 SXM | Too big | |||
| DGX Station GB300 | 1.6-bit, spills | 1M | 33 tok/s | 5.4 s |
| 8× H200 NVL | 2-bit | 1M | 58 tok/s | 0.9 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- llama.cpp
llama-server -hf unsloth/Kimi-K3-GGUF- vLLM
vllm serve moonshotai/Kimi-K3
Questions
How much VRAM does Kimi K3 need?
Kimi K3 needs 1,454 GB for its weights as released (INT4), 1,455 GB with an 8K context and 1,456 GB with 32K.
Can Kimi K3 run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the 8× H200 NVL.